What shipped · 2026-w46
What shipped in AI dev tools: week of Nov 9, 2026
Nov 9, 2026 – Nov 15, 2026
13 releases across 10 tools this week; biggest: Claude Code, GitHub Copilot, and Vercel AI SDK (2 releases each).
Claude Code
Claude Code 2.1.322
Released November 10, 2026. Fast mode is available on Vertex in the global, us, and eu regions. If Vertex refuses fast mode, the request falls back to standard speed.
Source for Claude Code 2.1.322 (opens in a new tab)Claude Code 2.1.321
Released November 9, 2026. Claude Fable 5.2 (claude-fable-5-2) is the default Fable model, with a 1M-token context window, $8 per million input tokens, $40 per million output tokens, and cache reads at $0.20 per million tokens. claude remote-control --settings applies a settings file to every session it starts, and if that file changes, no new session starts until claude remote-control is restarted. VS Code warns above the prompt when both an auth token and an API key are set. Saved Fable 5.1 model picks move to fable, on Pro and Max at once and on Team and Enterprise once the organization allows it. /model claude-fable-5-1 switches back. Mods Claude writes turn on without asking in auto and bypass modes, even with managed allowUnsandboxedCommands set to false or allowManagedDomainsOnly, and they do not load in dontAsk mode. A mod's own $.tool.call and $.mcp.call raise no permission prompt, even under ask rules or in dontAsk and claude -p runs. Deny rules and organization policy still apply. Fast mode labels in /fast, /config, and /model name the current model when it supports fast mode, and the VS Code row no longer says Opus only. With policy_mode set to layered, managed MCP servers add up over every matching gateway policy, including an empty match. Bedrock, Vertex, and Foundry upstreams ignore ANTHROPIC_*_BASE_URL variables in the gateway's environment. Set base_url on the upstream to use another endpoint. The release fixes MCP OAuth requests that sent secrets such as refresh tokens to another host when a server redirected them, a read-only Bash command that ran outside the sandbox without a permission prompt, Bash permission checks that judged a file write from the wrong folder, and deny rules and protected-file prompts that did not apply to the file a cp or mv into a directory writes. It also includes a large fix sweep for MCP, permissions, the shell, cloud sessions, Remote Control, VS Code, and the terminal UI.
Source for Claude Code 2.1.321 (opens in a new tab)
GitHub Copilot
Claude Fable 5.2 in GitHub Copilot
On November 9, 2026, Claude Fable 5.2 became generally available in GitHub Copilot for complex, multistep coding tasks across large codebases. GitHub says that in early testing it solved more tasks than Claude Fable 5.1 while using fewer steps and tool calls, and completed hard tasks that earlier models missed. It is available to Copilot Pro+, Max, Business, and Enterprise users in the model picker in Visual Studio Code, Visual Studio, Copilot CLI, the GitHub Copilot coding agent, the GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode, and Eclipse. Rollout is gradual. It is billed at provider list pricing under usage-based billing. Copilot Business and Copilot Enterprise administrators must enable the Claude Fable 5.2 policy in Copilot settings. The policy is off by default, and enabling it acknowledges that the model requires data retention. When it is used, Anthropic retains prompts and outputs to operate safety classifiers that detect harmful use. Retained data is not used to train Anthropic's models. Other Claude models, except Fable 5, Fable 5.1, and Fable 5.2, continue to operate under zero data retention. Eligible enterprises can use Fable 5.2, Fable 5.1, and Fable 5 with zero data retention through the end of the calendar year under a time-bound exemption while Anthropic rolls out Enterprise Frontier Safeguards. After that, continued use requires those safeguards. Approval does not enable the models. An administrator still has to turn the policy on. GitHub Support cannot determine eligibility.
Source for Claude Fable 5.2 in GitHub Copilot (opens in a new tab)GitHub Copilot CLI 1.0.108
Released November 9, 2026. settings.json adds sandbox.userPolicy.filesystem.enumeratePaths for Windows enumerate-only directory grants. /index commands are added, and managed shared tgrep indexing is on by default for eligible repositories. PDF attachments on Grok 4.7 and Grok 4.6 fall back to file references instead of failing the turn. With the sandbox on, a running local MCP server restarts after /cd so it follows the new directory's denied paths. An SDK file attachment that points at a named pipe, socket, or device is an error instead of stalling the turn. /model lists models when Copilot answers the model list with HTTP 421 by re-resolving the account endpoint. Ctrl+C stops a running /compact. MCP deny rules block a tool named the way its server advertises it, not only its sanitized name, even with --allow-all-tools. The notes say that since 1.0.106, Autopilot with Assisted Permissions no longer stops for permission prompts, and that since 1.0.104, Assisted Permissions judge calls are sent as agent requests and no longer count as separate premium requests. Managed plugin assignments with state defaultEnabled install and activate plugins by default while preserving user overrides. The release also includes a large fix sweep for the prompt, sandbox, MCP, sessions, and the terminal UI.
Source for GitHub Copilot CLI 1.0.108 (opens in a new tab)
Vercel AI SDK
Claude Fable 5.2 on AI Gateway
On November 9, 2026, Anthropic's Claude Fable 5.2 became available on Vercel AI Gateway. Vercel says it is built for open-ended technical work and improves on earlier Fable models in coding, reasoning, and vision. It can run long, ambiguous tasks unattended, and works as an orchestrator that hands simpler tasks to smaller models, or as an advisor that other Claude models consult. Effort runs from low to max, with high as the API default. Cybersecurity and biology safety classifiers are enabled. Access requires data retention. Anthropic does not support Zero Data Retention for the model. Prompts and completions are retained for 30 days and are not used to train Claude. If a team has Zero Data Retention turned on, AI Gateway returns an error instead of sending the request. A models array on providerOptions.gateway tries each model in order when the cybersecurity classifier declines a request. The model name is anthropic/claude-fable-5.2. Call it through the AI SDK, the Chat Completions, Responses, and Anthropic Messages APIs, or a coding agent connected to AI Gateway.
Source for Claude Fable 5.2 on AI Gateway (opens in a new tab)Muse Ember on AI Gateway
On November 9, 2026, Meta's Muse Ember became available on Vercel AI Gateway. Vercel describes it as a lightweight reasoning model for the high-volume parts of an agent run, such as coding sub-tasks, tool calls, and document analysis. It accepts text, image, and video input with a 1M-token context window. Tool calling is supported, and reasoning effort is adjustable from minimal to max. The model name is meta/muse-ember. Call it through the AI SDK, the Chat Completions, Responses, and Anthropic Messages APIs, or a coding agent connected to AI Gateway.
Source for Muse Ember on AI Gateway (opens in a new tab)
Anthropic API
Claude Fable 5.2 and Mythos 5.2
On November 9, 2026, Anthropic launched Claude Fable 5.2 (claude-fable-5-2) for demanding reasoning and long-horizon agentic work. Claude Mythos 5.2 (claude-mythos-5-2) offers the same capabilities only to organizations verified through Anthropic's verification programs, such as the Cyber Verification Program. Both models have a 1M-token context window, 128k max output tokens, and always-on adaptive thinking. They cost $8 per million input tokens and $40 per million output tokens. Claude Fable 5.1 and Claude Mythos 5.1 cost $10 and $50. Claude Fable 5.2 is available on the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed and Google managed), Claude in Amazon Bedrock, and Claude in Microsoft Foundry. Both models enforce the prefix check for every account. If code edits the system prompt, tools, or an earlier message and then sends back a Fable 5.2 or Mythos 5.2 thinking block from a later turn, the request returns a 400 error. On the Claude API and Claude Models on Google Cloud (Google managed), a conversation that moves onto either model keeps reasoning from other models, including Opus 5.5, Sonnet 5.5, and Haiku 5.5. Moving the other way drops the thinking blocks those two models produced, and the request still succeeds. Computer use uses the computer_toolset_20260801 toolset. A request that declares computer_20251124 returns a 400 error on every platform except Amazon Bedrock, where the earlier tool still works. The advisor tool is in beta. A Fable 5.2 or Mythos 5.2 executor accepts only one of those two models as its advisor, and every executor model accepts Fable 5.2 as its advisor. Both models require 30-day data retention and are not available under zero data retention unless Anthropic expressly authorizes it. Thinking blocks work only in the account that produced them, or in an account linked to it. Fast mode for Claude Fable 5.2 is available in Messages API requests on the Claude API. It delivers up to 2.5x higher output tokens per second, at $16 per million input tokens and $80 per million output tokens. Set speed to fast. No beta header is required.
Source for Claude Fable 5.2 and Mythos 5.2 (opens in a new tab)
Augment Code
Augment IntelliJ plugin 0.494.0
Released November 10, 2026. The notes cover changes since the last stable IntelliJ plugin, v0.491.0. The plugin supports JetBrains IDEs built on the 2026.3 platform (build 263). Long conversations keep more of their earlier context. The history limit counts message content rather than request metadata, and its default is larger. When older history is trimmed, it is dropped in larger steps instead of a little on every turn, so the context sent with each message stays consistent. Conversations save when many change at once or when one keeps failing to save. Failed saves are retried, and very large chat histories no longer abort saving or loading with a stack overflow. A chat response that stalls partway through ends with "The response timed out. Click retry to continue." and can be resumed with the existing retry flow. Connected tool integrations stay in Settings when a tool-status refresh after sign-in or authorization times out. A single agent turn no longer starts more than one background request to generate the conversation title.
Source for Augment IntelliJ plugin 0.494.0 (opens in a new tab)
Base44
SendEmail accepts HTML, plain text, and attachments
On November 9, 2026, Base44's SendEmail() integration accepts html as an alternative to body for HTML email, a text parameter for a plain-text version alongside the HTML, and an attachments array. Attachments can reference files the app already uploaded or inline base64-encoded content, with a limit of 5 files, 5 MB each, and 10 MB total.
Source for SendEmail accepts HTML, plain text, and attachments (opens in a new tab)
Devin Desktop
Devin Desktop 3.12.43
Released November 9, 2026. In Devin Local, a response that stops at the output limit before a tool call continues the turn, and GPT models no longer stop to ask for input they can work out. Image reads take an image_size. The default sends 70% of the original's longest side, at least 1024 px. Follow-up prompts in a Devin Local session are no longer blocked when Devin Cloud usage is restricted. Files in a gitignored directory are no longer listed as created. In the IDE, dragging a file or folder from the Explorer or an editor tab onto the Devin Local panel adds it as a chip. Child Devin Local sessions started in a Remote SSH window keep the parent summary. A Mermaid diagram inside a collapsed details section draws correctly when the section is opened, in Markdown previews and session messages. Saved codemaps no longer disappear on Windows. The Changes pane's Since last seen cutoff survives a tab switch and a restart. Mark unread is available in the sidebar row menu for the open session and for local sessions.
Source for Devin Desktop 3.12.43 (opens in a new tab)
Lovable
Jira, Mailjet, Sentry, and Resend connectors
On November 9, 2026, Lovable apps can use Jira and Mailjet connectors. Jira lets apps search, create, and update issues on a Jira Cloud site and read its projects, boards, and sprints. Mailjet lets apps send transactional email from a sender or domain validated in Mailjet, add contacts to lists, and read delivery status such as sent, opened, clicked, or bounced. The Sentry connector now works in the app as well as in the project chat. The app reports errors and performance data to a Sentry project with a client key (DSN). The Resend connector can connect by signing in with a Resend account instead of pasting an API key. Lovable refreshes that access automatically, and connecting with an API key stays available. Add a connector from Connectors, or ask Lovable in a project to connect it.
Source for Jira, Mailjet, Sentry, and Resend connectors (opens in a new tab)
Netlify
Muse Ember 1.0 in AI Gateway and Agent Runners
On November 9, 2026, Meta's Muse Ember 1.0 became available through Netlify's AI Gateway and Agent Runners. The model id is muse-ember-1.0, alongside the Muse Spark and Muse Wonder models already in AI Gateway. Netlify Functions can call it with the OpenAI SDK without a separate API key. The changelog example uses the Responses API. It also works in Scheduled Functions, Background Functions, and Edge Functions, and those calls use Netlify's caching, rate limiting, and authentication.
Source for Muse Ember 1.0 in AI Gateway and Agent Runners (opens in a new tab)
OpenCode
OpenCode 2.2.7
Released November 10, 2026. When a Copilot account has an Auto pool, Auto in the model picker lets Copilot's router choose the model for each session at the Auto discount. If routing fails, OpenCode falls back to the account's default chat model. Right-click a side panel tab to close tabs to the right. Command-W closes side panel tabs wherever focus is, Option-Command-W or Ctrl+Alt+W always closes the session tab, and Command-Shift-T reopens the last closed tab of either kind where it was. Mark as unread on a tab, Home row, project row, or inbox card keeps its dot until the next visit. Code that imports @opencode/ai directly reads response.images[0], because ImageResponse.image is removed. On a Location that is still loading plugins, agent.list returns the agents loaded so far, and integration.list, usage.limits, vcs.init, rpc.call, and artifact endpoints no longer wait up to 5 seconds. Model, provider, generate, voice, and websearch.query endpoints still wait. read accepts a negative offset, such as -50 for the last 50 lines. Remote MCP servers that rotate refresh tokens stay signed in when several clients refresh at once or the token endpoint fails for a moment. The release also includes a large fix sweep for MCP, the TUI, desktop, and the SDK.
Source for OpenCode 2.2.7 (opens in a new tab)
Updated
Newsletter
Get what shipped each week
One email when rankings change or a tool ships something that matters. No spam, unsubscribe anytime.