All 5 AI packs for $39 →
← All posts
News · Coding · Models

This week in AI tools: Claude Fable 5.2, Cursor live apps, OpenAI Images 2.6 and Kimi K3.1 Pro weights

Published · Facts checked against official vendor pages on November 10, 2026 · Rankings are independent of sponsorships

Anthropic's new top model arrived yesterday afternoon, after our Monday pricing roundup went out, so it leads this week. The rest of the week was about where agents do their work: Cursor's agents now deploy the apps they build and talk back to you, Copilot moved into Xcode and JetBrains Air, and OpenAI and Google both shipped faster image and video models. Every item below comes from the vendor's own announcement, changelog, docs or model card, with dates.

Tools in this post

The short version

Claude Fable 5.2 (Nov 9) costs $8 / $40 per 1M tokens, down from $10 / $50 for Fable 5.1, and Anthropic says it writes output up to 30% faster. Pro and standard Team seats get Fable included for a week, through November 16.

Fable 5.2 in GitHub Copilot (Nov 9) for Pro+, Max, Business and Enterprise, but admins have to switch on a policy that accepts Anthropic's data retention.

Cursor live apps (Nov 5) let a cloud agent build, host and share an internal app with Postgres and Cursor SSO. Voice chat (Nov 6) lets you talk to an agent while it works.

OpenAI Images 2.6 (Nov 4) is in ChatGPT, Codex and the API as Sunburst and Flare, with GIF and SVG output and markup-guided edits. Gemini Omni 1.2 Flash (Nov 5) makes 15-second clips at $0.07 per second of 720p.

Copilot in Xcode and JetBrains Air (Nov 3 and 4), and Kimi K3.1 Pro's open weights are now on Hugging Face.

Claude Fable 5.2: cheaper, faster, and still not your default

Fable 5.2 is Anthropic's model for the hardest, longest-running work. It has a 1M-token context, 128K max output and always-on adaptive thinking, and it is on the Claude API, AWS, Google Cloud and Microsoft Foundry as claude-fable-5-2. The list price fell 20%, to $8 per 1M input tokens and $40 per 1M output tokens. Fast mode, which Anthropic says is up to 2.5x faster, costs double ($16 / $80) and works in Claude Code and the API without a beta header. Its twin, Claude Mythos 5.2, has fewer cyber and biology safeguards and is only for organizations in Anthropic's verification programs.

Anthropic itself says to start most coding and knowledge work on Claude Opus 5.5 ($4 / $20) and step up to Fable only when a task needs it. Its own numbers show why the gap matters less than the sticker suggests: on its FrontierCode 2.0 eval, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort for roughly half the cost per task. Those are vendor benchmarks, so test on your own repos before you switch.

If you call the API directly, check three things before you move: Fable 5.2 rejects a request with a 400 if you edit the system prompt, tools or an earlier message and then send back a later thinking block; computer use needs the computer_toolset_20260801 toolset; and the model requires 30-day data retention, so it isn't available under zero data retention without Anthropic's say-so.

If you're on a Claude plan, this is the week to try it. Pro plans and standard Team seats get every Fable model included for one week, through November 16 at 11:59 pm PT, for up to 50% of their weekly limits, plus one Fable limit reset. Max, Ultra and premium Team seats get one to three resets based on how long they've paid, valid until March 8, 2027. Fable isn't picked for you; choose it in the model picker.

Fable 5.2 in Copilot and Cursor

GitHub Copilot made Fable 5.2 generally available on November 9 for Pro+, Max, Business and Enterprise, across VS Code, Visual Studio, the CLI, the coding agent, the Copilot app, github.com, mobile, JetBrains, Xcode and Eclipse, billed at provider list price. The catch is data handling. The Fable policy is off by default for Business and Enterprise, because Anthropic keeps prompts and outputs to run its safety classifiers (GitHub says not for training), while other Claude models in Copilot stay under zero data retention. Eligible enterprises can get a zero-retention exemption until the end of 2026 through their GitHub account team.

Cursor's models page, checked today, already lists Claude Fable 5.2 at the same $8 / $40 API list price. On Cursor's current plans, third-party models like Claude are billed from prepaid credit and need Premium ($30/mo) or higher. For a side-by-side of how these tools charge for premium models, see AI coding tools pricing compared and Cursor vs Claude Code vs Copilot.

Cursor agents now ship apps and take calls

Cursor's November 5 release goes after the apps that never leave localhost. Start a cloud agent, say you want to build an app, and Cursor scaffolds it, adds a Postgres database if needed, sets up sign-in behind Cursor SSO and provisions hosting. Publishing runs deterministic and agent-based security checks, for example for secrets left in app files, and nothing goes live until they pass and you click publish. Team admins can see every app and secret, add review rules and unpublish anything. Apps deploy from Cursor Origin or GitHub repos (GitLab is "coming soon"), are shared behind Cursor sign-in, and can be public on individual plans. It needs a paid plan with cloud agents, which on the current plans means Premium or above.

That puts Cursor closer to app builders like Lovable and Replit, but aimed at internal tools for people who already work in a repo. If you're choosing between those, our Lovable vs Bolt guide and What is vibe coding? explain the trade-offs.

On November 6 Cursor added voice chat with agents in the Agents Window, the IDE, the web and iOS. The voice agent hands coding work to the chat agent, and you can keep asking what changed in a diff or how a run is going while it works; the coding agent reads out short progress notes. It runs on Grok Voice and, per Cursor, with zero data retention when Privacy Mode is on. Dictation is a separate feature and still uses Cmd+Shift+Space.

Images 2.6 and Gemini Omni 1.2 Flash

OpenAI's Images 2.6 (Nov 4) is live in ChatGPT and Codex on every plan and in the API as two models. GPT-Image-2.6 Flare is the fast one: OpenAI says it can return a low-quality image in as little as six seconds and handles layouts for posters and product shots better. GPT-Image-2.6 Sunburst is for detail and for keeping a character or product consistent across edits. New in the API are a complexity setting (OpenAI suggests low for most edits), SVG and animated GIF output, and markup-guided edits, where you draw on the image to show what to change. See ChatGPT Images and OpenAI Codex.

Google's Gemini Omni 1.2 Flash (Nov 5) is now generally available on the Gemini API and Google Cloud at $0.07 per second of 720p video, 30% less than Omni 1.1. Clips run up to 15 seconds instead of 10, and you can change a clip turn by turn (a new background, a different product) instead of regenerating it. It is also rolling out in Flow, in the Gemini app for Google AI Ultra subscribers, and in Google Vids, where anyone with a Google account gets three free 15-second videos a month. See Gemini and Google Veo.

Copilot moves into Xcode and JetBrains Air

Since November 4, Copilot can be the engine behind Xcode's own coding assistant, in public preview from Xcode 27.3 beta on macOS Tahoe 26.6 or later, on every Copilot plan; set it up under Xcode > Settings > Intelligence. The separate Copilot for Xcode extension still exists. A day earlier, Copilot joined the agent picker in JetBrains Air, where cloud tasks run in a fresh container and bill against your Copilot plan rather than JetBrains AI credits. Air's own analytics won't show that Copilot usage, though, so track it on the GitHub side. Our free copilot-instructions-code-review and copilot-instructions-tests files work in either place. See JetBrains AI.

Smaller updates and deadlines

A few more items from official changelogs, plus two dates to put in your calendar:

  • Claude Console IP allowlists (Nov 5, beta): organizations can restrict API requests to up to 1,000 IPv4/IPv6 addresses or CIDR ranges, which limits where a leaked key works.
  • OpenAI released GPT-6-Cyber (Nov 5) for authorized vulnerability work through its Daybreak program, so it isn't a general API model. See ChatGPT.
  • November 30, 2026: fast mode for Claude Opus 4.8 is removed (requests return an error; Anthropic points to Opus 5.5), and Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) retires on the Claude API, with Sonnet 5.5 as the recommended replacement.
  • OpenAI says GPT-5.6 Sol's reduced $4 / $20 price runs at least through November 21, 2026.

Follow-ups from last week

Kimi K3.1 Pro weights: shipped. When we checked today, moonshotai/Kimi-K3.1-Pro was on Hugging Face, published about six days earlier, right around Moonshot's November 4 target, and updated November 5. It's a 2.8T-parameter Mixture-of-Experts model with 104B active parameters, a 1M-token context and MXFP4 weights. The Kimi K3.1-Pro License is permissive with two conditions: Model-as-a-Service companies with more than $20M in revenue over 12 months need a separate deal for commercial use, and products with more than 100M monthly users or $20M a month in revenue must show "Kimi K3.1-Pro" in their interface. See Kimi.

GPT-6.1 Terra in Cursor: still not listed. Cursor's models page today lists GPT-5.6 Sol, Terra and Luna but not GPT-6.1 Terra, which OpenAI released on October 26 and Copilot already offers.

Opus 4.8 fast mode: still scheduled for removal on November 30 in Anthropic's release notes. GitHub's model hosting page still lists Claude Opus 4.8 (fast mode) as a preview in Copilot, so if you use it there, plan your switch now.

What this means for our rankings

No tier changes this week. Fable 5.2 makes Claude and Claude Code stronger at the top end, but both are already S tier, and we change a tier only after a human review, not on launch day. Every item above has a dated, sourced entry in the changelog. If you're putting Fable or another agent to work on a real codebase, the Claude Code Agents pack and Agent Builder pack cover the config files these tools read, and the free CLAUDE.md, AGENTS.md and claude-code-reviewer files are a quick start.

FAQ

How much does Claude Fable 5.2 cost?
On the Claude API, Claude Fable 5.2 costs $8 per 1M input tokens and $40 per 1M output tokens, down from $10 / $50 for Fable 5.1. Fast mode, which Anthropic says is up to 2.5x faster, costs $16 / $80 per 1M tokens. Anthropic still recommends Claude Opus 5.5 ($4 / $20) as the starting point for most coding and knowledge work.
Can I use Claude Fable 5.2 on the Claude Pro plan?
For a limited time, yes. From November 9 through November 16, 2026 at 11:59 pm PT, Pro plans and standard Team seats get all Fable models included for up to 50% of their weekly usage limits, plus one Fable limit reset. After that week, Fable on those plans goes back to running on usage credits.
Is Claude Fable 5.2 available in GitHub Copilot?
Yes. GitHub made it generally available on November 9, 2026 for Copilot Pro+, Max, Business and Enterprise, billed at provider list price. Business and Enterprise administrators must enable its policy first, which is off by default because Anthropic retains prompts and outputs for Fable models to run safety classifiers.
Are Kimi K3.1 Pro's weights open?
Yes. Moonshot AI published Kimi K3.1 Pro on Hugging Face (moonshotai/Kimi-K3.1-Pro) around November 4, 2026. The Kimi K3.1-Pro License allows use, modification and redistribution, but large Model-as-a-Service businesses (over $20M revenue in 12 months) need a separate agreement for commercial use, and very large products must display "Kimi K3.1-Pro".

Free and paid rules for your agents

Related changelog

Sources

Prices change often. Always confirm on the vendor's site before buying. See the full tier lists and the changelog.

+ Your logo here