Summary
The dominant story this slot is the fallout from Anthropic pulling Claude Fable 5 and Mythos access, a cluster of roughly six posts spanning a class-action fraud suit over Max plan usage limits, rumors of ID verification gating a US-only return, and claims the shutdown traces to a US government directive. Riding alongside it, Kilo turned the Fable outage into a routing pitch: plan with your strongest model, implement with a cheaper one, the same service for 59% less, plus a separate note that MiniMax M3 was the only model to correctly call a soccer draw. A second cluster covers Grok Build's new multi-agent dashboard (three posts, including a Warp integration), and a small but real signal from Hugging Face and Cursor engineers dissecting the loss curve and learning-rate schedule of Mistral's next big model ("le chaton fat"). The rest is noise: @brivael posted 18 French political and SpaceX rants, and @AustinJustice, @spencerpratt, and Scobleizer's AlignedNews promo carry nothing AI-substantive.
Posts
- Anthropic Fable/Mythos shutdown saga (cluster of 6) (@ns123abc · WSJ). A federal class-action accuses Anthropic of marketing fraud, alleging the Max 5x/20x plans deliver far below advertised usage (one user hit 15% of weekly allowance in a single 5-hour sprint). Surrounding posts speculate Fable returns US-only behind ID verification, and joke the shutdown came from a government directive. Mostly rumor and meme, but the lawsuit and WSJ coverage are real.
- Kilo: plan with the strong model, implement with the cheap one (@kilocode · blog). Head-to-head test: Fable 5 wrote the better plan (9.1 vs 8.3), but both models passed all 15 acceptance checks when implementing the same plan, with GPT-5.5 costing $6.30 vs Fable 5's $16.66. The plan-strong/execute-cheap split is a concrete phase-level routing result. See llm-routing and the Kilo routing audit.
- MiniMax M3 calls the draw (@kilocode). Five models and one human predicted Qatar vs Switzerland; M3 was the only one to take the draw correctly, reportedly because it weighed form over the betting-odds line the benchmark excludes. Anecdotal but on-brand for the multi-model thesis. Relates to MaxProof MiniMax M3.
- Grok Build agent dashboard (cluster of 3) (@JasonBud, @ns123abc, @milichab). Grok Build adds a dashboard to manage multiple agents at once, with one user running ~5 agents in parallel. The Warp integration lets a SuperGrok subscription drive Grok models directly inside Warp.
- Mistral's next model loss curve ("le chaton fat") (cluster of 2) (@eliebakouch, @ellev3n11). HF and Cursor engineers note the learning-rate schedule matches DeepSeek V2's timing, except the second phase looks like a slight decay rather than constant. Hints at a pending Mistral tech report worth tracking.
- The rise of World-Action Models (@Scobleizer · NVIDIA blog). Moritz Reuss surveys WAMs (robot policies built on pretrained world-model or video backbones) as the second recipe for robot foundation models alongside classical VLAs. Robotics-adjacent, low priority, but a clean overview.
- Scobleizer AlignedNews / AhaCreator promo (@Scobleizer). Vendor demo of an influencer-marketing AI agent for his news brand. Skip.