$ git diff --stat

The Diff

Short notes when a tool, model or price changes something about how you ship with an agent. What changed, why it matters, and a link to the official source. Not a news feed: we only post when your workflow has to change. Written by MK. RSS

by tool: Kiro · v0 · Claude · Claude Code · OpenAI Agents API · OpenAI API · Claude API · Gemini API · Jules · GitHub Copilot · DeepSeek API · Codex · Anthropic · opencode · Lovable

Kiro adds Grok 4.7 with experimental support in the IDE, CLI, Crew and Web. 500K context, 2.7x credits for requests up to 200K tokens, 5.4x over that.

score credits past 200K tokens: 5.4x

  • Kiro changelog, dated November 10: Grok 4.7 is available with experimental support
  • Works in the Kiro IDE, CLI, Crew and Web
  • 500K token context window
  • 2.7x credits for requests up to 200K tokens, 5.4x over 200K
  • Affects Kiro Pro, Pro+, Pro Max and Power customers

open the article →

What changed

Kiro's changelog, dated November 10, adds Grok 4.7 with experimental support in the Kiro IDE, CLI, Crew and Web for Pro, Pro+, Pro Max and Power customers. The context window is 500K tokens. Requests up to 200K tokens are billed at a 2.7x credit multiplier; requests over 200K tokens, 5.4x. To see it in the model list, restart the IDE or CLI, run kirocrew restart for Crew, or refresh Kiro Web.

Why it matters if you ship with an agent

Credits past 200K tokens: 5.4x. The multiplier doubles at the line, and a long agent session crosses it without announcing itself. If you try it, watch the context size and compact before 200K. It is labelled experimental, so do not make it a team default yet.

sources: kiro.dev

open the article →

Kiro previews Enterprise Management: one AWS console policy for the IDE, CLI, Web and Crew LTS. On new us-east-1 profiles, models and API keys are denied until a rule allows them.

score regions in the preview: 1

  • Kiro changelog, dated November 10: Enterprise Management is in preview for newly created profiles in us-east-1
  • One policy applies in Kiro IDE 1.4.154 or later, CLI 3.1.0 or later, Kiro Web and Kiro Crew LTS
  • Models, API keys, and Crew messaging and network access are denied until a rule allows them
  • Files, shell, web, MCP and Crew scheduled jobs keep Kiro's defaults until a rule names them
  • Affects admins creating new Kiro profiles in us-east-1; existing and European profiles come later

open the article →

What changed

Kiro's changelog, dated November 10, puts Enterprise Management in preview for newly created profiles in US East (N. Virginia). Admins write one policy in the AWS console and it applies in the IDE (1.4.154 or later), the CLI (3.1.0 or later), Kiro Web and the new Kiro Crew LTS. Rules allow, deny or ask for models, shell commands, files, web search and MCP tools, for everyone, one group or one user. Models, API keys, and Crew messaging and network access are denied until a rule allows them. Files, shell, web, MCP and Crew scheduled jobs keep Kiro's defaults until a rule names them. Existing and European profiles come later.

Why it matters if you ship with an agent

Regions in the preview: 1. If your admin creates a new profile there, expect no models until a rule allows one. Update the IDE and CLI before you file the bug.

sources: kiro.dev

open the article →

v0 now works with ChatGPT Business plans, alongside Plus and Pro. Eligible responses from supported OpenAI models count toward the ChatGPT plan instead of v0 credits, up to a limit your admin can set.

  • v0's changelog, dated November 10, adds ChatGPT Business to the plans that work in v0, alongside Plus and Pro
  • Connect with Sign in with ChatGPT and eligible responses from supported OpenAI models count toward the ChatGPT plan instead of v0 credits
  • On Business a workspace admin can set usage limits for ChatGPT in other apps
  • At a limit v0 pauses generation and saves your progress, and you can ask the admin for more or continue on v0 credits
  • Affects teams on ChatGPT Business who build in v0, and the admins who set their limits

open the article →

What changed

v0's changelog, dated November 10, says ChatGPT Business plans now work in v0, alongside Plus and Pro. Connect with Sign in with ChatGPT, and eligible AI responses from supported OpenAI models count toward your ChatGPT plan instead of v0 credits. On Business, a workspace admin can set usage limits for using ChatGPT in other apps. Reach one and v0 pauses generation and saves your progress. From there you can copy a request asking the admin to raise the limit, or continue on v0 credits.

Why it matters if you ship with an agent

The bill moves. If your team already pays for ChatGPT Business, v0 work on OpenAI models can draw on that plan first. The entry does not list which models are supported or which responses are eligible, so check before you plan a budget around it. Admins: there is now a limit to set. Everyone else: a paused generation now means a message to your admin.

sources: v0.app

open the article →

Kiro IDE 1.4.154 adds Approve for me to chat. The Autopilot toggle becomes a three-mode menu, Autopilot is renamed Accept file changes, and a model can answer the prompts your rules leave open.

score modes in the menu: 3

  • Kiro IDE 1.4.154, dated November 10, adds Approve for me to chat
  • Where it is available the Autopilot toggle becomes a menu of Supervised, Accept file changes (formerly Autopilot) and Approve for me
  • Deny, ask and allow rules decide first, then a model approves, blocks or asks
  • Off by default, with a switch and a command to make it the default for new chats
  • Affects Kiro IDE users, and Kiro Web users, who got the same option on November 6

open the article →

What changed

Kiro IDE 1.4.154, dated November 10, adds Approve for me to chat. Where it is available, the Autopilot toggle becomes a permission mode menu: Supervised, Accept file changes (the old Autopilot) and Approve for me. Your deny, ask and allow rules decide first. For an action no rule covers, a model approves it if it judges it safe, blocks it if not, and asks when unsure. Each pass is marked Approved for you. The docs say it is off by default; "Use Approve for me by default" in What's new, or the command Kiro: Enable Approve For Me (User), turns it on for new chats. Kiro Web got the same option on November 6, under Agent Config, Tool approvals.

Why it matters if you ship with an agent

Modes in the menu: 3. Autopilot is gone by name, so update your team docs. Before making the third mode your default, write deny rules for push, deploy and secrets. Those still win.

sources: kiro.dev

open the article →

Claude Fable 5.2 and Mythos 5.2 are out at $8 in and $40 out per MTok, down from $10 and $50, with the prefix check enforced for every account.

score dollars off per MTok of input: "2"

  • Claude Fable 5.2 (claude-fable-5-2) launched November 9, with Mythos 5.2 for verified organizations
  • 1M token context, 128k max output, $8 input and $40 output per MTok, fast mode at $16 and $80
  • Prefix check applies to every account, and a thinking block sent after an edited prefix returns a 400
  • computer_20251124 returns a 400 except on Amazon Bedrock, and a 5.2 executor accepts only a 5.2 advisor
  • Affects anyone moving API code or agents from Fable 5.1 to 5.2

open the article →

What changed

Anthropic launched Claude Fable 5.2 (claude-fable-5-2) on November 9, with Claude Mythos 5.2 for verified organizations. Both have a 1M token context window, 128k max output tokens and always-on adaptive thinking, at $8 per MTok of input and $40 of output; the 5.1 models cost $10 and $50. Fast mode for Fable 5.2 costs $16 and $80. The prefix check now applies to every account, the earlier computer_20251124 tool returns a 400 everywhere except Amazon Bedrock, and a 5.2 executor accepts only a 5.2 advisor.

Why it matters if you ship with an agent

The model ID swap is the easy part. If your code edits the system prompt, tools or an earlier message and then sends back a later thinking block, the request returns a 400. Move computer use to computer_toolset_20260801, point the advisor at 5.2, and note that both models require 30-day data retention.

sources: platform.claude.com

open the article →

Claude Code 2.1.321 makes Fable 5.2 the default Fable model and moves saved 5.1 picks to it. Mods Claude writes now turn on without asking in auto and bypass modes.

score prompts raised by a mod's own tool calls: "0"

  • Claude Code 2.1.321, released November 9, adds Fable 5.2 as the default Fable model at $8/$40 per MTok with 1M context
  • Saved Fable 5.1 picks move to fable, at once on Pro and Max, on Team and Enterprise once the organization allows it
  • Mods Claude writes turn on without asking in auto and bypass modes and do not load in dontAsk mode
  • A mod's own tool calls raise no permission prompt, even under ask rules, while deny rules and organization policy still apply
  • Affects Claude Code users on Fable and anyone relying on ask rules to gate mods

open the article →

What changed

Claude Code 2.1.321, released November 9, adds Claude Fable 5.2 as the default Fable model: 1M context, $8/$40 per MTok. Saved Fable 5.1 picks move to fable, at once on Pro and Max plans and on Team and Enterprise once the organization allows it; /model claude-fable-5-1 switches back. Mods Claude writes now turn on without asking in auto and bypass modes, and do not load in dontAsk mode. A mod's own tool calls raise no permission prompt, even under your ask rules or in dontAsk and -p runs; deny rules and organization policy still apply.

Why it matters if you ship with an agent

Pro and Max users who saved Fable 5.1 are on 5.2 after the update, so check the model line before a long run. Ask rules no longer cover what a mod calls; anything a mod must never do belongs in a deny rule.

sources: github.com

open the article →

OpenAI's Agents API adds auto_review. Set it on a session and an OpenAI-managed model approves or denies shell, file, network and MCP requests. Off by default, tokens billed extra.

score per-tool overrides: 0

  • Session review field takes auto_review or none, default none
  • An OpenAI-managed reviewer model judges intent and risk, then approves or denies
  • Covers shell, file changes, extra permissions, network access and MCP confirmations, plus subagents
  • Browser sign-in, MCP forms and URL hand-offs still reach your app
  • Affects Agents API projects with access to the feature; review tokens billed on top

open the article →

What changed

OpenAI's API changelog for October 30 adds automatic review to eligible Agents API projects. Set a session's review field to auto_review and an OpenAI-managed model weighs each action's intent and risk, then approves or denies it. It covers shell commands and terminal input that need approval, file changes that need approval, requests for broader filesystem or network access, and MCP tool calls whose rules require confirmation, for the turn and its subagents. Browser sign-in, MCP forms and URL hand-offs still reach your app. Default is none; review tokens bill on top of the agent's usage, and a project without access gets the request rejected.

Why it matters if you ship with an agent

Who approves: a model the docs do not name. There is no per-tool override; what stays manual is what the environment forbids. Put network limits in the environment, not the reviewer, and read the turn's events afterwards.

sources: developers.openai.com

open the article →

Claude Code 2.1.319 adds permissions.denyInsteadOfPrompting, and auto mode with telemetry or feature flags off now asks or denies with no built-in classifier fallback.

score fallback classifiers: "0"

  • New setting denyInsteadOfPrompting denies any call that would wait on a prompt and the turn continues
  • Applies in terminal sessions, subagents and teammates
  • Auto mode with telemetry or feature flags off, or behind an LLM gateway, asks or denies when the server cannot check, no classifier fallback
  • CLAUDE_CODE_AUTO_MODE_SERVER=0 keeps the classifier, and claude -p now exits 1 when actions are refused
  • Affects unattended Claude Code runs, gateway users and anyone with telemetry off

open the article →

What changed

Claude Code 2.1.319, out November 6. A new permissions.denyInsteadOfPrompting setting denies any tool call that would wait on a permission prompt, and the turn continues, in terminal sessions, subagents and teammates. Auto mode with telemetry or feature flags off, including behind LLM gateways, now asks about or denies calls the server cannot check, with no fallback to the built-in classifier; CLAUDE_CODE_AUTO_MODE_SERVER=0 keeps the classifier, and Bedrock, Vertex and Foundry are unchanged. claude -p exits 1, not 0, when auto mode refuses actions because no safety check was available.

Why it matters if you ship with an agent

Unattended runs that hung on a prompt can now fail fast; turn the setting on and read the denials afterwards. Gateway users with telemetry off should expect prompts where auto mode used to decide. Scripts that watched claude -p for exit 0 get a 1 now.

sources: code.claude.com

open the article →

OpenAI added gpt-6-cyber to the API on November 5. $20 in, $100 out per million tokens, 1,050,000 context. Only orgs with Daybreak Red plus model approval can call it.

score approvals before the first request: "2"

  • Model id gpt-6-cyber, released November 5
  • For investigating vulnerabilities, reproducing them in authorized test environments and verifying fixes
  • $20 input, $2 cached input, $100 output per million tokens
  • 1,050,000 token context, 128,000 max output, Chat Completions and Responses only
  • Affects organizations approved for Daybreak Red; everyone else gets a refusal

open the article →

What changed

OpenAI's API changelog for November 5 lists GPT-6-Cyber, model id gpt-6-cyber, "to investigate vulnerabilities, reproduce them in an authorized test environment, and verify fixes." The model page prices it at $20 per million input tokens, $2 cached and $100 output: 1,050,000 token context, 128,000 max output, Chat Completions and Responses only. The Daybreak guide lists it as "Red + additional model approval": Daybreak Red access plus a separate approval for this model.

Why it matters if you ship with an agent

Most organizations cannot call it, so do not put the id in a config and hope. If you are approved, OpenAI's setup guide asks for a non-default project with Daybreak Red enabled, an IP allowlist for the addresses your app sends from, Private Safety Processing on any Zero Data Retention project, and a stable pseudonymous safety_identifier per user. Budget for output at five times the input rate.

sources: developers.openai.com

open the article →

Anthropic's November 3 release note says text from Claude Mythos 5, Opus 4.7 and Sonnet 4.6 now carries its text watermark; Sonnet 4.6 may skip it with custom sampling.

score hidden characters added: "0"

  • Anthropic's Claude API release note for November 3 says text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic's text watermark on the platforms it lists; that list includes the…
  • The docs say the watermark adds no tokens or hidden characters, carries no information about you or your organization, and needs no change to requests or response handling.
  • Nothing in your code has to change.

open the article →

What changed

Anthropic's Claude API release note for November 3 says text generated by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic's text watermark on the platforms it lists; that list includes the Claude API, the Claude apps, Claude Code, Cowork and Claude Tag. The docs say the watermark adds no tokens or hidden characters, carries no information about you or your organization, and needs no change to requests or response handling. On Sonnet 4.6, a request that sets temperature, top_p or top_k to a non-default value might not carry it. Detection is in private preview.

Why it matters if you ship with an agent

Nothing in your code has to change. If you pin Sonnet 4.6 and want the mark, drop all three sampling parameters, including any a framework sets for you. Need to check what Claude wrote? Detection is not open yet; register for the preview. Prose that goes out under your name now carries a mark a detector can read.

sources: platform.claude.com

open the article →

Claude Code 2.1.318 stops a flagged subagent instead of switching its model, no longer installs the IDE extension from an IDE terminal, and widens MCP deny lists.

score minutes before a background subagent's shell command is backgrounded: "3"

  • Claude Code 2.1.318, out November 4.
  • With "Switch models when a message is flagged" on "Ask each time", a subagent now stops instead of switching models.
  • An unattended run whose subagent gets flagged now stops rather than finishing on another model; check jobs that counted on the switch.

open the article →

What changed

Claude Code 2.1.318, out November 4. With "Switch models when a message is flagged" on "Ask each time", a subagent now stops instead of switching models. Starting in an IDE terminal no longer installs or updates the IDE extension; use /ide from another terminal. A deniedMcpServers URL entry now also blocks Anthropic-provided servers at that upstream URL, and connectors in cloud and self-hosted sessions. Agents created by /claude-api managed-agents-onboard get the auto permission policy by default. Background subagents move a shell command to the background after 3 minutes on all providers.

Why it matters if you ship with an agent

An unattended run whose subagent gets flagged now stops rather than finishing on another model; check jobs that counted on the switch. Audit MCP deny lists: a URL rule now takes Anthropic's own server at that address too. A long test run inside a background subagent is backgrounded at 3 minutes, so poll for the result. The extension stops updating itself from the terminal.

sources: code.claude.com

open the article →

Kiro CLI 3.0 starts new interactive sessions and --trust-all-tools scripts on V3, retires the Classic interface, the desktop app and five commands.

score commands retired: "5"

  • Kiro CLI 3.0.0, dated November 3, makes V3 the default.
  • New interactive sessions start on V3; existing V2 users are asked first.
  • The scripts flip quietly: a CI job that passes --trust-all-tools now runs on V3, and scripts are not asked first.

open the article →

What changed

Kiro CLI 3.0.0, dated November 3, makes V3 the default. New interactive sessions start on V3; existing V2 users are asked first. Headless scripts using --trust-all-tools start on V3, other scripts stay on V2, and new ACP clients start on V3. The Classic (V1) interface no longer runs: a command or setting that still selects it gets a "Continue with V3" offer. The desktop app is gone, and macOS installs switch to the standalone kiro-cli on their next update check. The autocomplete, inline, translate, theme and dashboard commands are unavailable. kiro-cli --v2 keeps the V2 agent.

Why it matters if you ship with an agent

The scripts flip quietly: a CI job that passes --trust-all-tools now runs on V3, and scripts are not asked first. Run it once by hand, or pass --v2 until you have. Grep your dotfiles for the five retired commands. Inline shell autocomplete ended with the desktop app and has no replacement.

sources: kiro.dev

open the article →

Claude Code 2.1.317 adds a check before a web or shell call sends your account email, and subagents on another model stop inheriting the session's effort.

score session effort levels your subagents inherit: "0"

  • Claude Code 2.1.317, released November 3, asks before a web or shell tool call sends your account email address to the network.
  • Subagents and teammates on a different model from the main conversation now use that model's saved effort level, or its default, instead of the session's effort level.
  • Headless runs are the thing to check.

open the article →

What changed

Claude Code 2.1.317, released November 3, asks before a web or shell tool call sends your account email address to the network. Subagents and teammates on a different model from the main conversation now use that model's saved effort level, or its default, instead of the session's effort level. Agent SDK hook callbacks can set onFailure: "block", so a callback that fails blocks the action instead of letting it run.

Why it matters if you ship with an agent

Headless runs are the thing to check. A claude -p job whose curl or git command carries your account email now hits a new check, and a headless job has nobody to answer it, so run those jobs once by hand after upgrading. If your session runs at high effort and your subagents on a cheaper model, those subagents just dropped to that model's own setting; set it per model if you were counting on the inherited one. SDK hook authors: a hook that errors still lets the action run unless you set the block option.

sources: code.claude.com

open the article →

Kiro CLI 2.37.0 upgrades V2 agent files to V3 when a V3 session starts unless you turn it off. Each file is rewritten in place after a .json.bak backup.

score backups per agent: 1

  • Kiro CLI 2.37.0, dated October 30, upgrades V2-format custom agents to V3 when a V3 session starts.
  • It runs unless you turn it off: an unset chat.enableAutoAgentUpgrade preference counts as on, and only an explicit false keeps it off.
  • Your agent files change the first time V3 starts, global ones included.

open the article →

What changed

Kiro CLI 2.37.0, dated October 30, upgrades V2-format custom agents to V3 when a V3 session starts. It runs unless you turn it off: an unset chat.enableAutoAgentUpgrade preference counts as on, and only an explicit false keeps it off. It covers your workspace and global agent directories, or the directory in KIRO_AGENT_CONFIG_DIR if you set one. Each agent file is rewritten in place after Kiro saves a .json.bak backup. To keep V2 agents, run kiro-cli settings chat.enableAutoAgentUpgrade false; interactive V3 launches then show a notice counting the agents still waiting. The same release lets ACP clients that support terminal authentication sign in from a terminal.

Why it matters if you ship with an agent

Your agent files change the first time V3 starts, global ones included. Diff each .json against its .json.bak before you trust the session. If the configs live in git, commit first, and set the preference to false on machines where the migration should land as a reviewed change.

sources: kiro.dev

open the article →

Kiro CLI 2.36.0 adds Approve for me to local V3 sessions. A classifier approves actions your deny, ask and allow rules leave open. Per workspace, off until you pick it.

score who approves: Sonnet 4.6

  • Kiro CLI 2.36.0, dated October 29, adds Approve for me to interactive local V3 sessions.
  • Run /permissions, choose Policies, then approve for me.
  • Who approves: Sonnet 4.6.

open the article →

What changed

Kiro CLI 2.36.0, dated October 29, adds Approve for me to interactive local V3 sessions. Run /permissions, choose Policies, then approve for me. Kiro approves actions it judges safe, blocks ones it judges unsafe and asks when unsure. Your deny, ask and allow rules decide first; the policy only handles what they leave open, and each action it passes is marked Approved for you. The docs say a Claude Sonnet 4.6 classifier makes the call, and the preset allows workspace reads, writes and read-only shell commands outright. It is stored per workspace, so other repos keep asking. The same release adds a subcommand match tier for shell rules, such as git stash *.

Why it matters if you ship with an agent

Who approves: Sonnet 4.6. Keep deny rules for push, deploy and anything touching secrets, since those still win. Then read the Approved for you labels after the session, not during it.

sources: kiro.dev

open the article →

Claude Code 2.1.316 deprecates fast mode on Opus 4.8 as of October 30, 2026. From November 30 those requests error with no fallback. Move to Opus 5.5.

score days of notice: 31

  • Claude Code 2.1.316 (October 31) deprecates fast mode for Claude Opus 4.8 as of October 30, 2026, with removal on November 30, 2026.
  • After removal, a fast mode request on Opus 4.8 returns an error and does not fall back to standard Opus 4.8.
  • If Claude Code runs Opus 4.8 with fast mode on, or a script sets speed to fast on claude-opus-4-8, it stops on November 30 with an error, not a slower answer.

open the article →

What changed

Claude Code 2.1.316 (October 31) deprecates fast mode for Claude Opus 4.8 as of October 30, 2026, with removal on November 30, 2026. After removal, a fast mode request on Opus 4.8 returns an error and does not fall back to standard Opus 4.8. The advice is to migrate to Opus 5.5 to keep the speedup. The fast mode docs carry the same warning.

Why it matters if you ship with an agent

Days of notice: 31. If Claude Code runs Opus 4.8 with fast mode on, or a script sets speed to fast on claude-opus-4-8, it stops on November 30 with an error, not a slower answer. Move those to Opus 5.5, where fast mode needs no beta header and the docs price it at $8 per million input tokens and $40 per million output, against $10 and $50 for Opus 4.8. Fast and standard speeds do not share the prompt cache, so watch the bill after the move.

sources: code.claude.com

open the article →

Google's Gemini API Computer Use tool is generally available on the Interactions and GenerateContent APIs. Paid tier only, billed as regular tokens.

score days until the recommended model doubles in price: 63

  • Google's Gemini API changelog for October 30: the Computer Use tool is now generally available across the Interactions and GenerateContent APIs.
  • The docs recommend gemini-3.8-flash for it, with gemini-4-argon and gemini-3.5-flash-lite also supported.
  • If a preview label was what kept your clicking agent off production, that reason is gone.

open the article →

What changed

Google's Gemini API changelog for October 30: the Computer Use tool is now generally available across the Interactions and GenerateContent APIs. The docs recommend gemini-3.8-flash for it, with gemini-4-argon and gemini-3.5-flash-lite also supported. It is paid tier only, billed as regular tokens at the model's rate. Gemini 3.8 Flash is $0.75 per million input tokens and $3.75 output through December 31, 2026, then $1.50 and $7.50 from January 1, 2027.

Why it matters if you ship with an agent

If a preview label was what kept your clicking agent off production, that reason is gone. Keep handling safety_decision: the model marks an action allowed, blocked, or needing confirmation, and for the last one you prompt the human and send safety_acknowledgement back. Prompt injection detection is opt-in and off by default, so switch it on before the agent reads a web page you did not write.

sources: ai.google.dev

open the article →

Max gets $100, Ultra $200 and Team up to $500 pooled in Claude API credits each billing cycle. Not for Claude Code, no rollover, claim after 7 days.

score months of rollover: 0

  • Anthropic's release notes (entry dated October 7): Max, Ultra and Team plans now include monthly Claude API credits.
  • Max gets $100, Ultra $200, and Team $20 per Standard seat and $100 per Premium seat, pooled and capped at $500.
  • Link the organization you actually build in: each plan links once, and only support can change it.

open the article →

What changed

Anthropic's release notes (entry dated October 7): Max, Ultra and Team plans now include monthly Claude API credits. Max gets $100, Ultra $200, and Team $20 per Standard seat and $100 per Premium seat, pooled and capped at $500. Claim them by linking one Claude Console organization under Settings > Billing on claude.ai; new subscribers wait 7 days. Credits refresh each billing cycle and expire unused, with no rollover. They cover the Claude API, Managed Agents, the Agent SDK and the playground. They do not cover Claude Code, extra usage in the Claude apps, or Claude on Bedrock, Google Cloud or Microsoft Foundry.

Why it matters if you ship with an agent

Link the organization you actually build in: each plan links once, and only support can change it. When the included balance hits zero and you hold no purchased credits, requests stop until the next cycle. Set a workspace spend limit or turn on auto-reload before a cron job finds out for you.

sources: platform.claude.com

open the article →

Google's Jules coding agent winds down on December 31, 2026. Sessions stop November 11, then it goes read-only. Export your settings and history first.

score days until read-only: 13

  • Google's October 29 changelog entry, titled Jules Turndown, says Jules will wind down on December 31, 2026, "as part of our work to unify our agentic coding efforts."
  • You have until November 11, 2026 to finish sessions.
  • Thirteen days to land anything half finished in Jules.

open the article →

What changed

Google's October 29 changelog entry, titled Jules Turndown, says Jules will wind down on December 31, 2026, "as part of our work to unify our agentic coding efforts." You have until November 11, 2026 to finish sessions. After that date Jules turns read-only until the end of the year. A new export saves everything inside Jules to a local folder: settings, environment variables, scheduled tasks, connected repositories, memories, setup scripts and every session with its prompt, plan, log and patches. You install Jules Tools, sign in, then paste Google's export prompt into a CLI agent such as Gemini CLI or Claude Code. Code and pull requests stay in GitHub. No successor is named.

Why it matters if you ship with an agent

Thirteen days to land anything half finished in Jules. Run the export before November 11, then move scheduled tasks and setup scripts to whatever agent you run next, because Jules is read-only after that date.

sources: jules.google

open the article →

GitHub Enterprise now bundles Copilot for every licensed user. No included allowance; usage is metered against the enterprise budget from request one.

score included allowance: 0

  • GitHub's October 28 changelog: GitHub Enterprise now includes Copilot for every licensed user, for enterprises with a GitHub or Microsoft account team.
  • Users without a GitHub Enterprise license can get Copilot Standalone at $10 per user per month.
  • If your company takes this model, the seat is free and every agent session hits the enterprise budget.

open the article →

What changed

GitHub's October 28 changelog: GitHub Enterprise now includes Copilot for every licensed user, for enterprises with a GitHub or Microsoft account team. Users without a GitHub Enterprise license can get Copilot Standalone at $10 per user per month. Both match Copilot Enterprise's features and models but carry no monthly AI credits per seat: usage is metered from the first request against the enterprise budget, with no included allowance. When the budget is gone, code completion and next edit suggestions keep working; Chat, code review and Copilot in the cloud stop. Copilot Business and Enterprise stay on sale. Available now for Enterprise Cloud; self-serve later, in stages.

Why it matters if you ship with an agent

If your company takes this model, the seat is free and every agent session hits the enterprise budget. Find out who sets it and what happens the week it runs dry: chat and cloud agents stop, autocomplete carries on.

sources: github.blog

open the article →

GitHub Copilot Fusion is generally available on all plans. It routes each prompt across models, bills each at its standard rate, and admins can make it the default.

score limited-time discount: 30%

  • GitHub's October 28 changelog makes Copilot Fusion generally available.
  • You pick Fusion in the model picker plus an effort level, Standard, High or Max, and it picks and chains the models per prompt: one model, a Cascade that drafts cheap and escalates, or a Critique where a second model…
  • Check your model picker: if you were on auto, you are on Fusion now, and your admin may have made it everyone's default.

open the article →

What changed

GitHub's October 28 changelog makes Copilot Fusion generally available. You pick Fusion in the model picker plus an effort level, Standard, High or Max, and it picks and chains the models per prompt: one model, a Cascade that drafts cheap and escalates, or a Critique where a second model reviews. It uses only models your enterprise and organization policies allow, and admins can set Fusion as the default model through Enterprise managed settings. Each model it uses is billed at that model's standard rate, with a 30% discount on paid plans for a limited time. Research preview users on auto or HydraFusion move over automatically. All plans, in VS Code, the Copilot app, Copilot CLI and JetBrains.

Why it matters if you ship with an agent

Check your model picker: if you were on auto, you are on Fusion now, and your admin may have made it everyone's default. One prompt can bill three models, so read the usage page before trusting the discount.

sources: github.blog

open the article →

Claude Code 2.1.305 stops allowedMcpServers filtering managed-mcp.json stdio servers that use ${VAR} only in env, so one you left out now loads on upgrade.

score allowlist rules that still block it: "0"

  • Claude Code 2.1.305, released October 20, changes how allowedMcpServers applies to managed-mcp.json.
  • Stdio servers that use ${VAR} only in env are no longer filtered by the allowlist: a server your allowlist left out loads on upgrade, and deniedMcpServers is the way to keep it off.
  • If an admin pushes managed-mcp.json and your allowlist was doing the filtering, a server you never approved loads after this upgrade.

open the article →

What changed

Claude Code 2.1.305, released October 20, changes how allowedMcpServers applies to managed-mcp.json. Stdio servers that use ${VAR} only in env are no longer filtered by the allowlist: a server your allowlist left out loads on upgrade, and deniedMcpServers is the way to keep it off. Plugins that Claude Desktop mounts by path no longer keep their allowed-tools when managed settings set allowManagedPermissionRulesOnly. A new sandbox.allowNestedUserNamespaces setting lets sandboxed commands on Linux create user namespaces, for Chromium and rootless podman, which lets them get around the sandbox's file protections.

Why it matters if you ship with an agent

If an admin pushes managed-mcp.json and your allowlist was doing the filtering, a server you never approved loads after this upgrade. Run claude mcp list after updating and add a deniedMcpServers entry for anything that should stay off. Leave the nested-namespaces setting alone unless a sandboxed command needs it; it opens the file protections by design.

sources: code.claude.com

open the article →

OpenAI's Responses API now offers 12-hour prompt caching on GPT-5.6 and later. Writes cost 2.5x the input price; the 30-minute default stays.

score cache write price: 2.5x

  • OpenAI's API changelog for October 20 adds a 12-hour prompt caching option.
  • It keeps shared instructions, tools and context cached so requests hours apart still get cached-input pricing.
  • Nothing changes until you set the option.

open the article →

What changed

OpenAI's API changelog for October 20 adds a 12-hour prompt caching option. It keeps shared instructions, tools and context cached so requests hours apart still get cached-input pricing. Per the prompt caching guide, you ask for it with prompt_cache_options.ttl set to 12h in the Responses API, on GPT-5.6 and later. The default stays 30m. A 12-hour cache write costs 2.5x the input token price, against 1.25x for the 30-minute write. Reads stay at 0.1x the input price on most of these models, and 0.05x on GPT-6.1 Sol.

Why it matters if you ship with an agent

Nothing changes until you set the option. It pays when the same long system prompt comes back after a gap: a nightly job, a scheduled agent, a reviewer that runs on each push. Writes cost twice as much, so count the gaps before switching. Upgrading an existing 30-minute entry charges a fresh 12-hour write, so set it from the first request, not after.

sources: developers.openai.com

open the article →

Kiro CLI 2.32.0 lets you allow a whole folder when a V3 session asks to write a file, runs cloud sessions non-interactively and previews workflow plans.

score files per yes: a folder

  • Kiro CLI 2.32.0, dated October 15: when a V3 session asks to write a file, you can now allow the whole folder instead of that one file.
  • V3 cloud sessions run from non-interactive commands.
  • The write prompt now has a bigger answer.

open the article →

What changed

Kiro CLI 2.32.0, dated October 15: when a V3 session asks to write a file, you can now allow the whole folder instead of that one file. V3 cloud sessions run from non-interactive commands. A workflow shows its plan before any step runs. The release also adds three V3 output styles and gives selected workflow steps access to Powers and MCP servers. Fixes cover non-interactive runs, model fallback, MCP sign-in and workflows.

Why it matters if you ship with an agent

The write prompt now has a bigger answer. Allowing a folder saves a tap per file, and it also means every later write under it goes through unchanged. Pick the narrowest folder that does the job, not the repo root, and keep anything with secrets or deploy config outside it. Workflow previews are the quieter win: read the plan before the first step, which is the one moment where saying no costs nothing.

sources: kiro.dev

open the article →

GitHub Copilot deprecated six models on October 19, 2026: GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash and Grok 4.5. Move to the named replacements.

score days of notice: 31

  • GitHub's changelog says six models are deprecated across most Copilot experiences as of October 19, 2026, as announced on September 18: GPT-5.5 and GPT-5.4 (use GPT-6.1 Sol), GPT-5.4 mini and GPT-5 mini (use GPT-6…
  • GitHub asks you to update it to a supported model.
  • If a Copilot workflow, agent mode setup or integration names one of these six, it stops working today rather than at some future date.

open the article →

What changed

GitHub's changelog says six models are deprecated across most Copilot experiences as of October 19, 2026, as announced on September 18: GPT-5.5 and GPT-5.4 (use GPT-6.1 Sol), GPT-5.4 mini and GPT-5 mini (use GPT-6 Luna), Gemini 3.7 Flash (use Gemini 3.8 Flash) and Grok 4.5 (use Grok 4.7).

Why it matters if you ship with an agent

If a Copilot workflow, agent mode setup or integration names one of these six, it stops working today rather than at some future date. GitHub asks you to update it to a supported model. Copilot Enterprise admins may need to enable the replacements under model policies in Copilot settings before they show up in the model picker in VS Code and on github.com. GitHub says no action is needed to remove the retired models themselves.

sources: github.blog

open the article →

DeepSeek upgraded V4.1 Flash to V4.1-Flash-1017 on October 17. The deepseek-flash model string, web and app all moved; price and API calls unchanged.

score model strings to update: 0

  • On October 17 DeepSeek upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017.
  • The model string deepseek-flash now returns the new model; the API calling method and pricing are unchanged.
  • If you call deepseek-flash, you are already on a different model and nothing in your config changed.

open the article →

What changed

On October 17 DeepSeek upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017. The model string deepseek-flash now returns the new model; the API calling method and pricing are unchanged. The official web and app models moved to it as well. DeepSeek says it scores higher across its benchmark table (Terminal-Bench 4.0 at 42.3, DeepSWE v1.1 at 77.0) with shorter task runtimes and shorter outputs, and asks DeepSeek Harness users to upgrade to 0.2.1-rc.1 or later. The same day it opened its Search / Fetch API, built on its in-house search engine.

Why it matters if you ship with an agent

If you call deepseek-flash, you are already on a different model and nothing in your config changed. Rerun your evals and golden prompts today, and check anything that depends on output length, since the new model writes less. Upgrade DeepSeek Harness before you blame the model.

sources: api-docs.deepseek.com

open the article →

Codex CLI 0.163.0, released October 15, restores a saved thread's permissions and network approvals on resume and drops the patched zsh shell backend.

score shell flags retired: 2

  • Codex CLI 0.163.0, released October 15: resuming a saved thread restores its permissions and network approvals, while respecting explicit overrides and current managed policy.
  • The network proxy now checks permission before a DNS lookup.
  • A thread you approved last week resumes with last week's approvals, not today's defaults.

open the article →

What changed

Codex CLI 0.163.0, released October 15: resuming a saved thread restores its permissions and network approvals, while respecting explicit overrides and current managed policy. The network proxy now checks permission before a DNS lookup. The patched zsh is gone: Codex uses its standard shell executor, and configurations that still set the retired shell flags, shell_zsh_fork and unified_exec_zsh_fork, get a migration notice. Daybreak, OpenAI's access programs for cybersecurity work, is enabled by default for eligible accounts.

Why it matters if you ship with an agent

A thread you approved last week resumes with last week's approvals, not today's defaults. Before you resume one, remember what you said yes to in it. Managed policy and explicit flags still win. If config.toml or a managed policy sets either zsh flag, it is ignored now; read the notice instead of scrolling past it.

sources: learn.chatgpt.com

open the article →

Fast mode is out of research preview for Claude Opus 5.5: no access request, no beta header, a default rate limit in the Start, Build and Scale tiers.

score beta headers required: 0

  • Anthropic's October 15 release notes take fast mode out of research preview for Claude Opus 5.5.
  • No access request and no fast-mode-2026-02-01 beta header; requests that still send it keep working.
  • Beta headers required: 0.

open the article →

What changed

Anthropic's October 15 release notes take fast mode out of research preview for Claude Opus 5.5. No access request and no fast-mode-2026-02-01 beta header; requests that still send it keep working. Opus 5.5 gets a default fast mode rate limit in the Start, Build and Scale tiers. Fast mode for Claude Opus 5 and Opus 4.8 stays in research preview and still needs the header. The docs price it at $8 per million input tokens and $40 per million output.

Why it matters if you ship with an agent

Beta headers required: 0. Set speed to fast on Opus 5.5 for up to 2.5x output tokens per second; the Evaluation tier has to move up to Start first. Keep the header for Opus 5 and 4.8. Fast mode has its own rate limit and switching speeds misses the prompt cache, so pick a speed per workload.

sources: platform.claude.com

open the article →

Politico reports Anthropic's draft bill would let a federal judge halt a frontier model only for imminent catastrophic risk, and preempt some state AI laws.

score bills introduced: 0

  • Politico reported Friday that Anthropic has privately shown lawmakers a draft frontier AI safety bill.
  • A model could be blocked only for a foreseeable, imminent risk of over $1 billion in damage or 50+ people killed or seriously hurt, from a bioweapon, loss of control, or a cyberattack.
  • Only Commerce, via a federal judge in DC, could suspend a model.
  • Anthropic says it reflects the Advanced AI Framework it has pushed since June.
  • Nothing changes this week.

open the article →

What changed

Politico reported Friday that Anthropic has privately shown lawmakers a draft frontier AI safety bill. It hasn't been introduced. As reported:

  • A model could be blocked only for a foreseeable, imminent risk of over $1 billion in damage or 50+ people killed or seriously hurt, from a bioweapon, loss of control, or a cyberattack.
  • Commerce-registered evaluators would test labs periodically. Labs could pay them, but not by result.
  • Only Commerce, via a federal judge in DC, could suspend a model. Some state AI safety laws would be preempted.

Anthropic says it reflects the Advanced AI Framework it has pushed since June.

Why it matters if you ship with an agent

Nothing changes this week. As described, the model you build on could only be pulled by court order over catastrophic risk, and one federal rule would replace parts of the state patchwork. Keep a fallback model wired in anyway.

More coverage

sources: politico.com · Reuters, via Yahoo News · Bloomberg

open the article →

Kiro retired Claude Sonnet 4 on October 14, 2026. Kiro does not switch your selected model for you, and requests to a removed model fail.

score models switched for you: 0

  • Kiro's changelog says Claude Sonnet 4 reached end of life in Kiro as of October 14, 2026, and that if your work depends on it you should choose another available model before continuing.
  • Claude Sonnet 5.5 is the latest Sonnet model.
  • If a Kiro agent, workflow or teammate's setup still picks Sonnet 4, nothing migrates it for you.

open the article →

What changed

Kiro's changelog says Claude Sonnet 4 reached end of life in Kiro as of October 14, 2026, and that if your work depends on it you should choose another available model before continuing. Claude Sonnet 5.5 is the latest Sonnet model. Kiro's model lifecycle page lists the same date for Claude Sonnet 4.0, marks EOL models [EOL] in the model picker, and says that after Kiro removes a model, requests to it fail. Kiro does not switch your selected model automatically.

Why it matters if you ship with an agent

If a Kiro agent, workflow or teammate's setup still picks Sonnet 4, nothing migrates it for you. Open the model picker and look for the [EOL] tag, then pick a model whose status is Active. Sonnet 5.5 is still listed as Experimental in Kiro and carries a 1.3x credit multiplier, so check the lifecycle page before you move everything there.

sources: kiro.dev

open the article →

GitHub Copilot deprecates Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra on November 16, 2026. Move to Claude Opus 5.5, GPT-6.1 Sol or GPT-6 Luna.

score days of notice: 32

  • GitHub's changelog says it will deprecate Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across all Copilot experiences on November 16, 2026: Copilot Chat, inline edits, ask and agent modes, and code completions.
  • The suggested alternatives are Claude Opus 5.5 for both Opus models, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra.
  • If an agent mode setup, a custom prompt or an integration names one of these three models, move it before November 16.

open the article →

What changed

GitHub's changelog says it will deprecate Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across all Copilot experiences on November 16, 2026: Copilot Chat, inline edits, ask and agent modes, and code completions. The suggested alternatives are Claude Opus 5.5 for both Opus models, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. One exception: Claude Opus 4.8 stays available to individual Copilot subscribers on annual plans, so they keep an Opus option.

Why it matters if you ship with an agent

If an agent mode setup, a custom prompt or an integration names one of these three models, move it before November 16. For Copilot Business and Enterprise, the alternatives are enabled automatically under default model enablement unless an admin turned off the global default or disabled the model. If so, enable them under model policies in Copilot settings. GitHub says no action is needed to remove the retired models themselves.

sources: github.blog

open the article →

Claude Code 2.1.300 makes remote-control --sandbox an error and validates hook-rewritten tool input: unknown keys dropped, rejected input denies the call.

score flags that ever worked: "0"

  • Claude Code 2.1.300, dated October 14, removes --sandbox and --no-sandbox from claude remote-control.
  • They had no effect; passing either now exits with an error, and sandbox.enabled in settings is how you sandbox a session.
  • Grep your launch scripts for remote-control --sandbox; that line now fails instead of being ignored.

open the article →

What changed

Claude Code 2.1.300, dated October 14, removes --sandbox and --no-sandbox from claude remote-control. They had no effect; passing either now exits with an error, and sandbox.enabled in settings is how you sandbox a session. The updatedInput a PreToolUse or PermissionRequest hook returns is now validated like Claude's own input: unknown keys are dropped, and input the tool rejects denies the call. Large requests to api.anthropic.com now use gzip even with telemetry off; CLAUDE_CODE_GZIP_REQUEST_BODIES=0 turns it off.

Why it matters if you ship with an agent

Grep your launch scripts for remote-control --sandbox; that line now fails instead of being ignored. If a hook rewrites tool input, a misspelled key is now dropped and a value the tool rejects denies the call, so run your hooks on a scratch repo before the next headless job. If a proxy between you and api.anthropic.com chokes on compressed bodies, set the gzip variable to 0.

sources: code.claude.com

open the article →

Claude Code 2.1.299 makes an mcpa allow rule match only the server a. Servers named ab or a_ need their own mcp * rule, in agent tools lists too.

score servers one rule reaches: "1"

  • Claude Code 2.1.299, dated October 14, narrows MCP permission rules.
  • An allow rule like mcpa now reaches only the server named a, not one named ab or a_.
  • A rule that was quietly covering a sibling server now prompts for its tools, and a non-interactive run cannot answer a prompt.

open the article →

What changed

Claude Code 2.1.299, dated October 14, narrows MCP permission rules. An allow rule like mcp__a now reaches only the server named a, not one named a__b or a_. To allow those, write mcp__<server>__* with the server's whole normalized name. Agent tools: lists take only the named server's tools, with the same spelling for names that contain __. The release also fixes deny and ask rules that were not applying to servers with a double underscore in their normalized name.

Why it matters if you ship with an agent

A rule that was quietly covering a sibling server now prompts for its tools, and a non-interactive run cannot answer a prompt. Check every mcp__ rule in settings.json and agent frontmatter against claude mcp list, and respell the ones for servers with __ or a trailing _ in the name. Deny rules on those servers may only have started working today.

sources: code.claude.com

open the article →

Kiro CLI 2.30.0 adds a startup prompt to switch to V3. V3 becomes the default harness with CLI 3.0 at the end of October, and V3 now shows model fallback notices.

  • Kiro CLI 2.30.0, dated October 12, shows a startup prompt inviting you to switch to V3.
  • It is rolling out gradually, so you may not see it yet.
  • The prompt migrates your custom agents in one tap, and the default flips in about three weeks either way.

open the article →

What changed

Kiro CLI 2.30.0, dated October 12, shows a startup prompt inviting you to switch to V3. It is rolling out gradually, so you may not see it yet. Choosing "Switch to V3 and upgrade my configs" saves V3 as your default harness, turns on automatic agent upgrades and starts migrating your custom agent configurations. V3 becomes the default with CLI 3.0 at the end of October. You can try it any time with kiro-cli --v3. In V3 sessions, when a model refuses a request or is unavailable and the turn moves to another model, the transcript now shows a notice.

Why it matters if you ship with an agent

The prompt migrates your custom agents in one tap, and the default flips in about three weeks either way. Run kiro-cli --v3 on a scratch branch first and check that upgraded agents kept the access they had on V2. Then read the fallback notices: a refusal or an outage can hand your turn to a different model mid-session, and now you can see when it did.

sources: kiro.dev

open the article →

Google made gemini-4-argon generally available on October 13 at $2 in and $10 out per million tokens through November 30, then $4 and $20. Sampling params are ignored.

score price on December 1: 2x

  • On October 13 Google made Gemini 4 Argon, model id gemini-4-argon, generally available in the Gemini API and AI Studio.
  • Input, thinking and output share one window of 1,048,576 tokens.
  • Seven weeks at half price, then the meter doubles.

open the article →

What changed

On October 13 Google made Gemini 4 Argon, model id gemini-4-argon, generally available in the Gemini API and AI Studio. Input, thinking and output share one window of 1,048,576 tokens. Pricing is $2 per million input tokens and $10 output through November 30, then $4 and $20 from December 1. Cached input goes from $0.10 to $0.20. It is not on the Free tier or the Google AI Pro plan. The model ignores temperature, top_p and top_k, and explicit caching is not supported. GitHub Copilot added it the same day for Pro+, Max, Business and Enterprise.

Why it matters if you ship with an agent

Seven weeks at half price, then the meter doubles. Run your evals at low, medium and high thinking and keep the lowest that passes. Set max_output_tokens to at least 32,768, since thinking spends the same budget, and client timeouts to 15 minutes. Delete any caches.create code: only implicit caching works, from 4,096 tokens up.

sources: ai.google.dev

open the article →

opencode v1.19.0 no longer installs updates by itself. Run /update in the TUI, or opencode upgrade --major to move to OpenCode 2.

score updates installed for you: 0

  • opencode v1.19.0, published October 11, stops the TUI from installing updates automatically.
  • The release notes say to use /update to install updates or to move to OpenCode 2, and that updates are no longer installed automatically.
  • Updates installed for you: 0.

open the article →

What changed

opencode v1.19.0, published October 11, stops the TUI from installing updates automatically. The release notes say to use /update to install updates or to move to OpenCode 2, and that updates are no longer installed automatically. The same release adds opencode upgrade --major as the command that upgrades to OpenCode 2. The notes do not link to OpenCode 2 release notes or a migration guide.

Why it matters if you ship with an agent

Updates installed for you: 0. If you relied on opencode patching itself, it stopped. Pick a cadence and run /update yourself, or pin the version in your setup script. OpenCode 2 is a major version and the upgrade is now a manual command, so check what changed in it before running opencode upgrade --major on a machine that ships code. Scripts and CI images that assumed the latest build need an explicit update step.

sources: github.com

open the article →

GitHub Copilot's JetBrains plugin now needs JetBrains IDE 2025.2 or later. Support for 2025.1 ended with the October 10 release.

score days of notice: 0

  • GitHub's October 10 changelog for Copilot in JetBrains IDEs says Copilot has ended support for JetBrains IDE 2025.1.
  • You now need JetBrains IDE 2025.2 or later to use the plugin.
  • If any machine is pinned to IntelliJ, PyCharm, WebStorm or another JetBrains IDE at 2025.1, Copilot is no longer supported there.

open the article →

What changed

GitHub's October 10 changelog for Copilot in JetBrains IDEs says Copilot has ended support for JetBrains IDE 2025.1. You now need JetBrains IDE 2025.2 or later to use the plugin. The same release adds an MCP setting to turn off automatic server startup for Copilot and Claude, and lets enterprise administrators set the default Copilot agent model for new conversations through managed settings. Users can still pick a model themselves.

Why it matters if you ship with an agent

Zero days of notice. If any machine is pinned to IntelliJ, PyCharm, WebStorm or another JetBrains IDE at 2025.1, Copilot is no longer supported there. Move the IDE to 2025.2 or newer before the next plugin update. On an enterprise plan, ask whether an admin set a default model, because new chats may open on something other than the one you were using.

sources: github.blog

open the article →

Google deprecated gemini-3.7-flash and gemini-3.5-flash on October 8. Every request to them now runs on gemini-3.8-flash or gemini-3.6-flash.

score days of notice: 0

  • On October 8 Google deprecated two Gemini API models. gemini-3.7-flash is replaced by gemini-3.8-flash, and gemini-3.5-flash by gemini-3.6-flash.
  • All requests to the old ids are automatically routed to the replacements, so anything still naming them already runs on a different model.
  • Grep for gemini-3.7-flash and gemini-3.5-flash in configs, env files and whatever model picker your agent calls.

open the article →

What changed

On October 8 Google deprecated two Gemini API models. gemini-3.7-flash is replaced by gemini-3.8-flash, and gemini-3.5-flash by gemini-3.6-flash. All requests to the old ids are automatically routed to the replacements, so anything still naming them already runs on a different model. Neither entry gives a shutdown date, and the deprecations page says none is announced. The pricing page now lists only the replacements: $0.75 per million input tokens and $3.75 output on both, through December 31, 2026, then $1.50 and $7.50 from January 1, 2027.

Why it matters if you ship with an agent

Zero days of notice. Grep for gemini-3.7-flash and gemini-3.5-flash in configs, env files and whatever model picker your agent calls. Change the strings to what you are actually getting, then rerun your evals, because the model under them changed yesterday without a commit. Lovable and Copilot are already retiring 3.6 Flash, so 3.8 is the longer road. Check the 2027 price before you settle in.

sources: ai.google.dev

open the article →

Google deprecated the deep-research-pro-preview-12-2025 agent on October 8 and shuts it down on October 23, 2026. Two April 2026 previews replace it.

score days of notice: 15

  • On October 8 Google deprecated the Deep Research Agent deep-research-pro-preview-12-2025 in the Gemini API and set its shutdown for October 23, 2026.
  • Two newer agents replace it: deep-research-preview-04-2026, which Google positions for speed and streaming results back to a client UI, and deep-research-max-preview-04-2026, for maximum comprehensiveness in automated…
  • Grep for the old agent id in anything that calls the Interactions API, including the research step an agent wired up in January.

open the article →

What changed

On October 8 Google deprecated the Deep Research Agent deep-research-pro-preview-12-2025 in the Gemini API and set its shutdown for October 23, 2026. Two newer agents replace it: deep-research-preview-04-2026, which Google positions for speed and streaming results back to a client UI, and deep-research-max-preview-04-2026, for maximum comprehensiveness in automated context gathering and synthesis. Migration is a one-field change: the agent parameter in interactions.create requests.

Why it matters if you ship with an agent

Fifteen days. Grep for the old agent id in anything that calls the Interactions API, including the research step an agent wired up in January. Pick the April preview that matches the job: the plain one when a person is waiting on the stream, max when a pipeline consumes the output. Then rerun whatever consumes the report, since the two replacements are tuned differently.

sources: ai.google.dev

open the article →

Codex CLI 0.161.0, released October 7, sets GPT-6.1 Sol as the default in the bundled and Amazon Bedrock catalogs, and widens what an approved filesystem escalation can write.

score days from release to default: 8

  • Codex CLI 0.161.0, released October 7, makes GPT-6.1 Sol the default model in the bundled and Amazon Bedrock catalogs.
  • OpenAI released gpt-6.1-sol on September 29 at $2 in and $10 out per million tokens for prompts up to 272K input tokens.
  • Type /model: if you never pinned one, you are now on GPT-6.1 Sol.

open the article →

What changed

Codex CLI 0.161.0, released October 7, makes GPT-6.1 Sol the default model in the bundled and Amazon Bedrock catalogs. OpenAI released gpt-6.1-sol on September 29 at $2 in and $10 out per million tokens for prompts up to 272K input tokens. The models page calls GPT-6 Sol the previous Sol model. The same release lets an approved filesystem escalation grant broader write access while keeping denied reads and network restrictions, and background tasks keep the permissions of the turn that started them.

Why it matters if you ship with an agent

Type /model: if you never pinned one, you are now on GPT-6.1 Sol. Pin it in config.toml if the old behaviour was the point. On Enterprise and Edu it stays off until an administrator enables it, and Free and Go are not included at launch. Then re-read your approval prompts. "Approve" on a filesystem escalation covers more than it did last week.

sources: learn.chatgpt.com

open the article →

Anthropic launched claude-haiku-5-5 on October 7 at $0.10 in and $0.50 out per Mtok under 100K tokens. Haiku 4.5 thinking budgets now return a 400 error.

score more tokens, same text: 30%

  • On October 7 Anthropic released Claude Haiku 5.5, model id claude-haiku-5-5: 1M token context, 128k max output, adaptive thinking on by default.
  • Prompts up to 100,000 tokens cost $0.10 per million input tokens and $0.50 output.
  • Grep for budget_tokens, temperature and top_p next to any haiku model string.

open the article →

What changed

On October 7 Anthropic released Claude Haiku 5.5, model id claude-haiku-5-5: 1M token context, 128k max output, adaptive thinking on by default. Prompts up to 100,000 tokens cost $0.10 per million input tokens and $0.50 output. Over that, $0.50 and $2.50. Haiku 4.5 is $1 and $5. Claude Code 2.1.293 makes it the default Haiku. Code written for Haiku 4.5 can break: budget_tokens, temperature, top_p, top_k and assistant prefill now return a 400 error, and the same text counts as about 30% more tokens.

Why it matters if you ship with an agent

Do not just swap the id. Grep for budget_tokens, temperature and top_p next to any haiku model string. Recount prompts and raise max_tokens values tuned for 4.5, since thinking tokens count against them. Pick content blocks by type, not position. Prompts over 100K tokens pay five times the rate. Claude Code's /claude-api migrate runs Anthropic's ten-step guide for you. Read what it changes.

sources: platform.claude.com

open the article →

Lovable deprecated google/gemini-3.6-flash for app AI features on October 6. It stops working there on November 19, 2026. Gemini 3.8 Flash replaces it.

score days of notice: 44

  • On October 6 Lovable marked Gemini 3.6 Flash, google/gemini-3.6-flash, deprecated for its apps' AI features, saying Google is retiring it.
  • Apps that use it keep working until November 19, 2026, then it stops.
  • If a Lovable app names the model, search the project for gemini-3.6-flash and change it to gemini-3.8-flash.

open the article →

What changed

On October 6 Lovable marked Gemini 3.6 Flash, google/gemini-3.6-flash, deprecated for its apps' AI features, saying Google is retiring it. Apps that use it keep working until November 19, 2026, then it stops. Lovable points to Gemini 3.8 Flash, google/gemini-3.8-flash, which it says accepts the same input at the same price. Google's own Gemini API deprecations page lists no shutdown date for gemini-3.6-flash as of today. GitHub Copilot deprecated the same model across all Copilot experiences on October 2, also pointing to Gemini 3.8 Flash.

Why it matters if you ship with an agent

If a Lovable app names the model, search the project for gemini-3.6-flash and change it to gemini-3.8-flash. Lovable's AI connector now lists every model a feature can use as a card, so open it and check what each feature actually runs on. Then test those features. Same price is Lovable's claim. Same answers is not.

sources: docs.lovable.dev

open the article →

Anthropic deprecated claude-sonnet-4-5-20250929 on September 30, 2026. Requests to it fail after November 30. The replacement is claude-sonnet-5-5.

score days of notice: 61

  • On September 30 Anthropic deprecated Claude Sonnet 4.5, model id claude-sonnet-4-5-20250929, and set its retirement on the Claude API for November 30, 2026.
  • Deprecated means it still answers but is no longer recommended.
  • Grep your repos for the id.

open the article →

What changed

On September 30 Anthropic deprecated Claude Sonnet 4.5, model id claude-sonnet-4-5-20250929, and set its retirement on the Claude API for November 30, 2026. Deprecated means it still answers but is no longer recommended. Retired means requests to it fail. The recommended replacement is claude-sonnet-5-5. The dates apply to the Claude API, the Claude Platform on AWS and Microsoft Foundry. Amazon Bedrock and Google Cloud set their own schedules.

Why it matters if you ship with an agent

Grep your repos for the id. An agent config, subagent definition or CI script written last autumn has a fair chance of pinning it, and the agent that pinned it will not warn you when it stops. The Usage page in the Claude Console exports a CSV of calls by API key and model, which finds the ones your repo does not. Swap to claude-sonnet-5-5 and rerun whatever tests cover the output. That is 54 days from today.

sources: platform.claude.com

open the article →

OpenAI's API now has three paid usage tiers, Build, Launch and Grow, unlocked automatically at $5, $100 and $500 of total credit purchases.

score tiers removed: 2

  • On October 6 OpenAI collapsed the API's five paid usage tiers into three.
  • Build unlocks at $5 in total credit purchases and allows $500 of usage a month.
  • If your Codex or app traffic runs on an API key, open the organization's limits page and see which of the three names it landed on.

open the article →

What changed

On October 6 OpenAI collapsed the API's five paid usage tiers into three. Build unlocks at $5 in total credit purchases and allows $500 of usage a month. Launch unlocks at $100 and allows $5,000 a month. Grow unlocks at $500 and allows $200,000 a month. An organization moves up on its own as its total credit purchases reach each threshold. The Free tier still exists in allowed regions, capped at $100 a month, and per-model rate limits still rise with tier.

Why it matters if you ship with an agent

If your Codex or app traffic runs on an API key, open the organization's limits page and see which of the three names it landed on. The monthly cap is what stops an agent loop halfway through a task. The $500 step to Grow is the one that matters: that is where the large limits live.

sources: developers.openai.com

open the article →

Google deprecated gemini-3.1-flash-image on October 6 and shuts it down on October 29, 2026. Nano Banana 2.1 is the replacement and is GA now.

score days of notice: 23

  • On October 6 Google made gemini-nano-banana-2.1, an update to Nano Banana 2, generally available.
  • Google says it keeps Flash-level speed and cost and improves visual quality, prompt adherence, multi-turn character consistency, text rendering and wide and panoramic ratios (1:4, 4:1, 1:8, 8:1) at 1K, 2K and 4K.
  • Grep your repo for the old model id.

open the article →

What changed

On October 6 Google made gemini-nano-banana-2.1, an update to Nano Banana 2, generally available. Google says it keeps Flash-level speed and cost and improves visual quality, prompt adherence, multi-turn character consistency, text rendering and wide and panoramic ratios (1:4, 4:1, 1:8, 8:1) at 1K, 2K and 4K. The same entry deprecates gemini-3.1-flash-image and sets its shutdown for October 29, 2026.

Why it matters if you ship with an agent

Grep your repo for the old model id. Anything still calling gemini-3.1-flash-image stops working on October 29, and the agent that picked that id in July will not warn you. Swap it to gemini-nano-banana-2.1, rerun whatever tests cover image output, and check sizes and ratios in the results, since the wide ones are new.

sources: ai.google.dev

open the article →

The weekly roundup lives in the Vibe Log.