Model intelligence.
Follow the changes. Understand what comes next.
Why AI Router ZDR Labels Are Not a Guarantee: Inside OpenRouter's MiniMax Warning and a 4-Step Hardening Guide
Following Dragos's November 19, 2026 study of two million canary-seeded prompts, OpenRouter emailed affected customers on November 20–21 warning that prompts and outputs sent to MiniMax models from July to October 2026 may have been read by a third party—whether served directly or forwarded via AtlasCloud, Novita, or StreamLake—and removed ZDR labels from affected endpoints pending investigation.
Latest stories126
Newest publishedWhy the Claude API Now Rejects temperature, and How to Fix the 400 Error
On October 21, 2026, Anthropic released Claude API version 2026-10-21. Under this version, a request returns a 400 error if it sends temperature, top_p or top_k (even at the default value), places a document block directly in an assistant message, or sends a text block with a non-empty citations array in a user message.
How to Direct Characters, Camera and Pacing in Kling 4.0
Kling 4.0 launched October 15, 2026. Kuaishou says it generates up to 30 seconds natively in one pass, combines up to 15 reference assets and accepts up to 10 keyframes. This guide maps those controls to characters, camera moves and pacing, and lists what Kuaishou has not announced yet, including prices, plans and regions.
What Is Claude Duet? Who Can Get It and How the Waitlist Works
Anthropic opened a research preview of Claude Duet on October 21, 2026. It is a standalone app where Claude acts as a proactive agent running on Claude Opus 5.5. It is open to Pro and Max subscribers in the US through a waitlist, at no extra cost, and it shares the plan's usage limits. Background checks count toward those limits. Team and Enterprise plans can't use it yet.
Best Coding Agents Right Now, Ranked by Artificial Analysis (October 2026)
On the Artificial Analysis Coding Agent Index v1.5, three setups share the top spot: Claude Code with Claude Sonnet 5.5 (max), Devin Fusion CLI with Claude Opus 5.5 (max) plus GPT-6 Luna (high), and Muse Code with Muse Wonder Full (max), all between 68.0 and 68.4. This page groups the full leaderboard into tiers, lists each agent's model and setting, and gives token-based API cost per task.
Kimi Code updated? Check the skill name before reinstalling the plugin
A plugin skill that stops responding after an update can look like a broken installation. With Kimi Code CLI 2.3.0, there is a simpler possibility to check first: the name used to invoke it has…
Did Claude Opus 5.5 Get Nerfed? What the Data Shows So Far
Since late September, users on Reddit, X and GitHub have said Claude Opus 5.5 got worse after launch and drains usage limits faster. BridgeBench's third-party NerfBench has kept every Opus 5.5 run inside its normal-variance band, and its October 16 run put the model at 105.8% of launch. Anthropic hadn't responded publicly as of October 21, Beijing time.
Which Models GitHub Copilot Offers Right Now, and What Each One Costs
A plan-by-plan list of GitHub Copilot models, how AI Credits pricing works, and which models retired on October 2 and October 19 or retire on November 16.
Here's How Gemini 4 Argon Usage Is Counted in the Gemini App, Antigravity and the API in October 2026
Google doesn't publish a set number of Gemini 4 Argon prompts for any plan. In the Gemini App, usage is compute-based and refreshes every five hours until a weekly limit. In Antigravity, Google AI Pro gets a separate weekly-only Argon quota and Ultra gets the highest limits, with no published figures. On the Gemini API, Argon is paid-only, and limits depend on the project's usage tier. This guide, based on Google pages as of 00:40 Beijing time on October 21, explains each one and what to do when you reach a limit.
Gemini CLI Counts Requests per Day, and Here's How Many Each Sign-In Method Gets
Gemini CLI limits how many model requests you can make each day, and the number depends on how you sign in. Google's Gemini CLI quotas page as of 22:42 Beijing time on October 20, lists 1,000 requests a day for a free Google account, 250 for a free Gemini API key, 1,500 for Google AI Pro and Gemini Code Assist Standard, and 2,000 for Google AI Ultra, Gemini Code Assist Enterprise and Workspace AI Ultra. This guide covers how those limits are counted, what Gemini CLI does when you reach one, and the ways Google documents to get more.
Here's What Every Claude Model Costs on Anthropic's API in October 2026
Anthropic's pricing page lists four current Claude models on its API: Claude Fable 5.1 at $10 per million input tokens and $50 per million output tokens, Claude Opus 5.5 at $4 and $20, Claude Sonnet 5.5 at $2 and $10, and Claude Haiku 5.5 from $0.10 and $0.50. This guide covers those prices, the older models still on the price list, and the modifiers that change a bill, including prompt caching, the Batch API, long prompts on Haiku 5.5, US-only inference, fast mode and tool charges. All figures are as of 22:38 Beijing time on October 20, 2026, per Anthropic's pricing page and API docs.
Gemini CLI Says "429 Resource Exhausted"? Here's What Google's Docs Say and What to Try
Gemini CLI users who run into a 429 see an error that Google's Gemini CLI FAQ quotes as "API error: 429 - Resource exhausted." Google's documentation says it means a request limit has been exceeded, and it lists what to do next: wait and retry, check your usage, let the CLI fall back to another model, switch to an API key or Vertex AI, upgrade to a higher quota or request an increase. This article covers what the error looks like, what each Google document says it means and the steps those documents give.
Claude Code Says "529 Overloaded"? Here's What Anthropic's Docs Say and What to Try
When Claude Code can't get past a 529 overloaded_error, it stops with "API Error: Repeated 529 Overloaded errors." Anthropic's documentation says the API is temporarily at capacity, that a 529 is not your usage limit, and that checking the status page, waiting a few minutes or switching models with /model are the things to try. This article covers what the message looks like, how Claude Code retries before showing it and what the docs suggest next.