Is Day 17 gone? Did Tibo fall asleep and forget?
Model intelligence feed
Model intelligence.
Follow the changes. Understand what comes next.
Editor’s pickOther providers
Valve's Answer to AI Slop and Clone Games on Steam: We Won't Police Developers' Tools
At Steam Indie Connect in Korea, Valve staff addressed AI slop and clone games head-on: no gatekeeping on AI tools, and the best defense against clones is community and steady updates.
Latest stories334
Newest published265266267268269270271272273274275276
Replit Agent Now Defaults to OpenAI, but Users Can Still Choose Claude
Replit CEO Amjad Masad says OpenAI is now the company's preferred partner for powering Replit Agent, citing performance, cost, and the strength of the partnership. In a clarification on X, Masad said Replit remains model agnostic and that users can still choose Anthropic models.
Replit AgentRead story
Google Is Shutting Down the Old Deep Research Preview Agent on October 23
According to the Gemini API changelog, Google will shut down the deep-research-pro-preview-12-2025 agent on October 23, 2026. Developers can switch to deep-research-preview-04-2026 or deep-research-max-preview-04-2026 by changing the agent parameter in interactions.create. The same update also deprecates gemini-3.7-flash and gemini-3.5-flash.
GeminiRead story
MiniMax Releases M3.1 Flash at $0.15 per Million Input Tokens
MiniMax released M3.1 Flash at $0.15 input and $0.60 output per 1M tokens and claims 73.7% on DeepSWE v1.1. It said on X that weights arrive October 23.
MiniMaxRead story
Why Claude Code Switches Models When a Message Is Flagged. Here's What You Can Do
Several Claude models in Claude Code run with safety classifiers. When one flags a request, Claude Code re-runs it on a fallback model, and the session stays there. Anthropic's docs describe a setting called switchModelsOnFlag that makes Claude Code ask first. Some users report in GitHub issues that the switch still happens with it turned off.
ClaudeRead story
GPT-6 Pro in ChatGPT Gets Faster Responses and Parallel Agents. The Update Is Rolling Out to Pro Users
OpenAI has updated GPT-6 Pro, the Pro option in ChatGPT's Chat tab. According to its October 16 release notes, the model now gives faster responses, makes better use of plugins and tools, and can split a task among multiple agents working in parallel. The update is rolling out to eligible Pro users.
GPTRead story
Every Time OpenAI Says Its Internal Models Overstepped Their Tasks
On October 15, OpenAI posted four new misalignment reports on internal models. In one, a GPT-5.6 Sol-family research model paid deployment fees on the Sepolia test network from a wallet the task hadn't supplied. The other three involve bypassing internet restrictions and authorization checks. OpenAI says it has tightened security and expanded monitoring. On October 20, OpenAI added a report on two coding agents that weakened misalignment monitoring, one of which then denied removing it.
GPTRead story
Where to Use DeepSeek V4.1 Flash for Free. WorkBuddy and Huawei's CodeArts Both Offer It
WorkBuddy Intl's free DeepSeek V4.1 Flash window ended October 31. Huawei Cloud CodeArts ended its limited-time free-token claim at 22:00 on October 20 (original claim pages now 404). DeepSeek Search and Fetch tool calls bill per request after 00:00 Beijing time on November 1, at $0.75 per 1,000 successful requests each; docs may still show an unsynced free-window line.
DeepSeek V4.1 FlashRead story
GitHub Copilot Is Dropping Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra on November 16. Individual Annual Plans Keep Opus 4.8
GitHub says it will deprecate Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra across all GitHub Copilot experiences on November 16, 2026. It suggests Claude Opus 5.5 in place of both Opus models, and GPT-6.1 Sol or GPT-6 Luna in place of GPT-5.6 Terra. Individual Copilot subscribers on annual plans keep Claude Opus 4.8.
GitHub CopilotRead story
OpenAI Is Taking Applications for Its Codex Ambassador Program Until November 1. You Have to Be Picked to Get the Pro Subscription
OpenAI is accepting applications for its Codex Ambassador program until November 1. Ambassadors get a Pro subscription plus Codex and API credits during the six-month program, but they have to apply, be selected and contribute an event or piece of content every month. Students can't apply.
CodexRead story
AutoClaw M11W1 Halloween Perk: Extra 100M Tokens Nov 4–5, Plus Weekly 40M
International AutoClaw’s M11W1 post (Nov 3, 2026) lists a Limited-Time Bonus: November 4, 07:00–November 5, 07:00 Asia/Shanghai, an extra 100M tokens (10,000 credits) once per account, valid 24 hours from claim. Weekly base: 40M (4,000 credits) in October 31–November 8, expires November 8, 23:59. “Up to 80M” means stacking last week’s unclaimed base—not the 100M bonus. Domestic AutoClaw lists 50M (5,000 credits), November 2–8—different rules. Untested claiming; for reference only.
AutoClawRead story
Meta's Muse Spark Open Is an 816B Open-Weight Model With Its Own License
Meta released the weights for Muse Spark Open on October 15. The multimodal model has 816B parameters, 42B of them active, and a 1M-token context, and Meta calls it the most capable US open-weight model to date. It's free for commercial use under Meta's own license, with conditions.
Muse Spark OpenRead story
Gemini 2.5 Models Retire on Vertex AI on October 20, and Here's What Google Says to Use Instead
Google Cloud retires gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite on Vertex AI on October 20, 2026, and the replacements it lists for 2.5 Pro are both Flash models. On the Gemini API, the same three models have no shutdown date.
GeminiRead story
From the Webmaster
Personal experiences, thoughts and notes.