← Dailies

Daily · 2026-11-10 00:00 UTC

Anthropic's Fable 5.2 retakes the top, Google adds two TPUs

Anthropic ships Claude Fable 5.2 and Mythos 5.2 to first place on Artificial Analysis, Google expands its eighth-generation TPU line to four systems in a year, and Disney names Adobe its official AI innovation partner.

Source ↗

1. Anthropic's Claude Fable 5.2 and Mythos 5.2 take first place — The two models share a foundation but differ in access: Fable keeps safeguards that block risky cybersecurity and biology tasks and is open to all users, while Mythos lifts some of those restrictions and is limited to organizations verified through Anthropic's Cybersecurity or Life Sciences programs. Pricing is $8 per million input tokens and $40 per million output tokens, down from $10 and $50, and output is up to 30 percent faster than the 5.1 versions. On Artificial Analysis's Index v5 at the highest compute tier, Fable 5.2 scores 65 points — five ahead of Gemini 4 Carbon and 15 ahead of Fable 5.1 — at an average of $26.50 per task. The Decoder · Nov 9
2. Google adds TPU 8tx and 8i Max, its fourth TPU system this year — TPU 8tx is aimed at secure, confidential inference and TPU 8i Max at massive context windows with the highest compute density per rack. It is the first time Google has introduced four TPU systems in a single year, following 8t for large-scale training and 8i for real-time reasoning, which Google said delivered up to 2.7 times better training performance and 80 percent better inference price performance than the previous generation. Google frames the additions around agentic workloads, where regulated customers want encryption and agents need to reason across long contexts. Google · Nov 9
3. Meta releases Muse Ember, a cheap model built for agent grunt work — Meta Superintelligence Lab's lightest and most cost-efficient model yet is priced at 10 cents per million input tokens and targets the work agents repeat all day: tool calls, coding subtasks, routing, extraction and long document parsing. It handles text, images and audio, and is designed to run underneath a larger planner — as Meta puts it, "Muse Wonder plans, Muse Ember handles the volume." Ember is the fourth Muse model, after Spark in April, its open-source twin Glimmer in August and the Wonder frontier model in October. TechCrunch · Nov 9
4. ByteDance's Seedream 5.5 promises 100 rounds of editing without drift — The image model is built for production workflows: ByteDance says unchanged areas stay pixel-identical across up to 100 editing rounds with no color shifts, and the model outputs native RGBA images with transparent backgrounds, so generated elements work as design assets without cutouts. ByteDance also claims it leads GPT Image 2.5 Sunburst on dense small text and realism. Seedream 5.5 is rolling out through Jimeng, Doubao, Doubao Work and Volcengine Ark, with a Pro version planned. The Decoder · Nov 9
5. Anthropic buys $3.4 billion of gas turbines — Anthropic is spending $3.4 billion on Caterpillar natural gas generators, buying units that equipment supplier OnePWR had ordered years earlier but could not finance, according to a Financial Times report. The generators can produce up to 1.7 gigawatts. It is a shift for a company that has preferred to rent: its May S-1 disclosed paying SpaceX subsidiary xAI $1.25 billion per month through May 2029 for compute. Turbine backlogs now stretch into the early 2030s, which leaves little room to wait. TechCrunch · Nov 9
6. Apple starts selling $50 Intelligence Usage Packs — Logic Pro now offers a "Buy More Usage" option at a one-time $49.99, for Apple Intelligence features that hit limits because they rely on cloud models from OpenAI or Google Gemini. The packs are tied to the purchasing Apple Account — no Family Sharing — roll unused usage month to month, and expire 12 months after purchase. Only Logic Pro has the option so far; Pixelmator Pro, Final Cut Pro, Motion, Pages, Keynote and Numbers are expected to get it this month. The Verge · Nov 9
7. Palantir pushed a minister to extend its AI into NHS GP records — WhatsApp messages obtained under freedom of information by campaign group Foxglove, and shared with the Guardian, show exchanges between Palantir's UK boss Louis Mosley and former health minister Zubir Ahmed spanning July 2025 to March 2026. Palantir argued "the logic of joining up primary and secondary [care] data is unarguable" and that its software was "ideal" for the job, despite GP data being expressly ruled out of the scope of its £330m contract to run NHS England's federated data platform. The British Medical Association and patient groups oppose bringing GP data onto the platform. The Guardian · Nov 9
8. Google ties school search access to its classroom AI — Google said schools that block its AI tools will also lose access to Google Search, and that it is updating its approach to controls for Gemini, Gemini Notebook, Gemini in Classroom and Search. The company says schools should have control over whether students use AI in the classroom. Parents and teachers told the Guardian the policy pushes districts toward adopting tools they distrust; one called the threat "nothing but extortion." The Guardian · Nov 9
9. Nominal raises $240M at a $3B valuation to put an agent on hardware testing — The Series C was led by Lightspeed Venture Partners and Insight Partners, with Lux Capital, Sequoia, General Catalyst, Founders Fund, Avenir, Salesforce Ventures, M13 and Lockheed Martin Ventures also participating. The round triples Nominal's valuation to $3 billion, up from $1 billion set in March with an $80 million round. Its platform pulls together sensor readings, logs, video and simulation output from hardware tests, and its agent, Nominal Analyst, investigates the data — used across aircraft, rocket engines, autonomous vehicles, drones and energy systems. SiliconANGLE · Nov 9
10. Disney names Adobe its official AI innovation partner — The multi-year alliance, announced at Adobe MAX, gives Disney's creative teams custom generative models built on Adobe Firefly Foundry, trained on Disney characters, environments and visual styles. Walt Disney Imagineering's research and development arm is already using a sketch-to-image model that turns hand-drawn concepts into rendered 2D concept art, plus a custom image model for on-brand franchise assets, with StudioLAB and ESPN Edge projects underway. Both companies describe the output as "human-centered and artist-led" and "commercially safe." GamesBeat · Nov 9

🧪 Try This Next

• 🔧 Split your agent's work: let a strong planner set the rules and a cheap model execute the repetitive steps — that is the exact split Meta is selling with Wonder plus Ember.
• 🎥 Watch Adobe MAX's Firefly Foundry demos, then test Seedream 5.5's multi-round editing on one real asset to see where drift actually shows up.
• 📝 Audit which AI features you use have hidden usage limits, and what topping them up would cost — Apple just turned that into a $49.99 product decision.

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.