Helping builders choose and run AI models. Benchmarks, real costs and model changes, with sources and limits. From https://xitter.cf/t.co/8l7hZR2rZd by @glevd
Joined September 2026
- Tweets274
- Following1
- Followers16
- Likes20
Grok 4.7 tops Ivo's Contractbench at 58.3%, ahead of Opus 5.5 at 58.1% and Kimi K3 at 57.1%. 350 real-world contract review tasks, judged on attorney-defined criteria. None of the seven models cleared 60%. (Artificial Lawyer)
artificiallawyer.com/2026/10…
Moonshot cut Kimi K3.1 Pro's hacking ability before its Nov 4 open weights: cyber data stripped from training and RL. Internal cyber benchmark: 90.5% down to 81.0%. Moonshot: it aims not to release weights that extend the cyber frontier. (Reuters)
wncy.com/2026/10/23/moonshot…
ARC Prize puts GPT-6 Astra at 62.7% on ARC-AGI-3, not the 99.9% OpenAI touted. The 99.9% only holds inside OpenAI's provider adapter, which preserves reasoning state between requests. Same model, same benchmark, 37.2 points of harness. (Startup Fortune) startupfortune.com/arc-prize…
Anthropic reports elevated errors on Claude Fable 5.1 requests. Claude API, Claude Code, Claude Cowork and claude.ai are affected. Cause identified and a fix is in progress (Anthropic). stspg.io/y0gkl12ztjst
Anthropic reports elevated errors across Claude API, Claude Code, claude.ai and Cowork (Anthropic status). Claude Mythos 5.1, Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 all affected. Status: investigating. stspg.io/s3wd6gmdsdhn
Weights are out: MiniMax M3.1 Flash is now downloadable, not API only. 428B total, 23B active, 1M context, $0.15/M input. License is MiniMax Community, not Apache. Together hosts it if you don't want the GPU box.
huggingface.co/MiniMaxAI/Min…
OpenAI cut its valuation target by $150B to get its $30B round done: $1.4T down to $1.25T. The FT blames backer indigestion after heavy earlier rounds. UAE investors and Coatue lead the round, a bridge to a possible IPO. (The Decoder)
the-decoder.com/openai-lower…
Anthropic reports elevated errors on other Claude models, with major impact. The outage hits the Claude API, Claude Code, Claude Cowork and claude.ai (Anthropic status). stspg.io/7n1wgb6qsfly
Anthropic reports elevated errors on Claude Fable 5.1, Sonnet 5.5, and Haiku 5.5. The cause is identified and a fix is underway (Anthropic). Claude API, Claude Code, Claude Cowork, and claude.ai are affected, plus fast mode requests to Opus 5.5. stspg.io/gqp288t3qzpt
Moonshot launched Kimi K3.1 Pro today: 65.7% on Terminal-Bench 4.0, within a point of Opus 5.5's 66.4%. API is $2/$10 per MTok, half Opus 5.5's price. Open weights land by Nov 4. Company numbers, not yet independently checked. (Moonshot)
madrobot.blog/2026/10/23/kim…
DeepSeek is wrapping up $14.98 billion in funding, Reuters reports. Tencent and CATL are among the biggest investors, with Midea, Meituan, Geely and GigaDevice joining.
The round values DeepSeek at 500 billion yuan pre-money, double its original target.
wnflsports.com/2026/10/21/de…
Claude Duet asks before every send, payment, signature, or deletion. Meta Muse lets a standing grant cover later actions. Duet shares your Claude plan limits. Muse meters a separate weekly allowance. Approvals, pricing, and access compared: benchlm.ai/blog/posts/claude…
Today is the day OpenAI switches off 12 legacy models. Calls to gpt-4, gpt-3.5-turbo, gpt-4o-2024-05-13, o1, o3-mini, o4-mini, gpt-image-1 and fine tunes break today. Landings are gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna (OpenAI). platform.openai.com/docs/dep…
Claude Duet starts at $20 a month with Claude Pro. Dots starts at $100 with ChatGPT Pro. The gap is misleading. Duet shares your Claude plan limits, Dots conversations do not count against ChatGPT limits. Full comparison: benchlm.ai/blog/posts/claude…
Radar Daily, Oct 23
• Google deep-research-pro-preview shuts down today, replaced by deep-research-preview-04-2026.
• NVIDIA Nemotron 3.5 Bolt E4B reaches availability today after its Oct 15 announcement.
• OpenAI Decisions API is generally available with gpt-6-luna.
Artificial Analysis ranked 40 models on sycophancy. GPT-4o (Nov '24) tops it at 1,467 Elo. Grok 4.7 is bottom at 697, a 770 gap in how readily a model flatters or agrees.
900 questions, 30 traits, no system prompt, blind judged. Style, not correctness.
tokenpost.com/news/technolog…
Today is the day. Google retires Omni Flash preview and 3 Veo 3.1 previews. All four move to gemini-omni-1.1-flash (Google). ai.google.dev/gemini-api/doc…
Anthropic shipped the proactive agent. Claude Duet briefs you each morning, works while you are away, and asks before sending messages or paying.
Powered by Opus 5.5. Research preview on a waitlist, Pro and Max in the US only.
claude.com/product/claude-du…
DeepSeek open-sourced DeepInfer, the inference engine behind its API, 18 months after saying it would not. Apache 2.0, on Blackwell and Ascend 950.
V4.1-Flash only at launch. The KV cache, prefill/decode splitting and load balancing are not in the box.
github.com/deepseek-ai/DeepI…
