Key Points
- Meta launched Muse Ember on Monday 9 November, its smallest and cheapest Muse model, built for the repetitive work AI agents do all day.
- Anthropic launched Claude Fable 5.2 and Claude Mythos 5.2 the same day, aimed at demanding coding, reasoning and agent work.
- Ember costs $0.10 per million input tokens and $0.50 per million output tokens. Fable 5.2 costs $8 and $40, eighty times more per token at the top end.
- Fable 5.2 is open to everyone. Mythos 5.2 is restricted to vetted organisations in Anthropic’s cybersecurity and life-sciences verification programmes.
- Anthropic says Fable 5.2 generates output up to 30 percent faster than its predecessor and can coordinate more than 1,000 AI agents at once.
- Both launches land as competition between Meta, Anthropic, OpenAI and Google intensifies ahead of Anthropic’s planned stock market listing.
Meta and Anthropic launched new AI models on the same day, Monday 9 November, at opposite ends of the market: one to make everyday agent work cheap, the other for the hardest technical jobs.
What happened
Meta introduced Muse Ember, the smallest and most efficient model in its Muse family, and the fourth model under the Muse name. It is built for the jobs AI agents repeat all day: calling tools, running small coding tasks, routing requests, pulling data out of documents and reading long files. It can take in a million tokens at once, write up to 512,000 tokens in reply, and work with text, images, audio and video. It went live on Monday through the Meta Model API and Muse Code, under the model name muse-ember-1.0, and it is also listed on OpenRouter. The price is $0.10 per million input tokens and $0.50 per million output tokens, with cached input at $0.004. Meta is keeping the model’s weights private.
Meta says Ember delivers more intelligence per dollar than Anthropic’s Claude Haiku 5.5 or OpenAI’s GPT-6 Luna at every reasoning level. On the Artificial Analysis Intelligence Index it scores 36 at its highest reasoning setting, matching Haiku 5.5’s best score, at about $0.14 per task against Haiku’s $0.41. Meta’s AI chief, Alexandr Wang, said it offers nearly the intelligence of the company’s top model, Muse Spark 1.3, at 17 times lower cost per task. Ember is also the first model Meta has trained using Muse Wonder Full, its frontier model from October. The comparison is not all in Ember’s favour: on real coding work in the FrontierCode 2.0-AA test, Haiku scored 34.9 percent against Ember’s 27.2 percent.
Anthropic, meanwhile, released Claude Fable 5.2 and Claude Mythos 5.2, the latest versions of its most capable model line. The two are the same underlying model, split apart by safety rules. Fable carries safeguards that block most cybersecurity and biology requests, which is what lets Anthropic offer it to everyone. It is available now in Claude’s paid plans, Claude Code, the Claude Platform and the major cloud services, Amazon Web Services, Google Cloud and Microsoft Azure, and coding-tool maker Cognition put it in Devin Desktop and Devin CLI on launch day. Mythos has those safeguards switched off and is limited to organisations accepted into Anthropic’s Cyber Verification Program or Life Sciences Verification Program, plus enterprise customers through Claude Security.
Both models cost $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for the 5.1 versions. Anthropic says they generate output up to 30 percent faster than their predecessors and communicate more clearly. A fast mode, at $16 and $80 per million tokens, generates tokens up to 2.5 times faster. Anthropic says the models improve coding, reasoning and visual work, and it still recommends its Opus 5.5 model as the starting point for everyday coding and knowledge work, reserving Fable and Mythos for the most ambitious technical projects.
Anthropic backed the launch with striking research claims of its own. It says a team of 1,000 Fable 5.2 agents working together turned the existing proof of Fermat’s Last Theorem into machine-checked form in 30 hours, with little human help. That is a formalisation of the centuries-old proof, not a new proof of the theorem. The company also says the model worked through two open problems in theoretical black-hole physics, pointing to preprints, and designed three electronic devices that worked without circuit fixes. On its own internal test for finding known software flaws, Mythos 5.2 found 10.2 percent more vulnerabilities than Opus 5.5. None of these results has been checked by outside researchers yet, so they stand as Anthropic’s claims for now.
Why Muse Ember and Fable 5.2 matter
The two launches, hours apart, show the AI market splitting in two directions at once. At the bottom, the contest is about cost: agents perform hundreds of small tasks, and the price of each one decides whether agent products can make money. Ember’s pitch is that a model scoring 36 on a leading index can cost $0.14 a task while the flagship model beside it costs $2.39 for a near-identical score. At the top, the contest is about capability: Fable 5.2 is aimed at work no smaller model can touch, from coordinating more than 1,000 agents at once to steering robots and hardware.
The gap between the two price tags is the story in one number. Fable 5.2’s output tokens cost eighty times Ember’s. That is not a mistake or a mismatch; it is two companies aiming at different buyers. Meta wants its model inside every agent workflow that runs thousands of times a day. Anthropic wants its model on the hardest jobs a company has, where a single answer can be worth far more than the tokens it took to produce.
The timing is no accident either. Reuters reported that OpenAI, Google and Meta are all releasing new models aimed at coding, reasoning and agents to win over developers and business customers, and Google added two new TPU systems on the same day to power exactly this kind of work. Anthropic is also heading for a planned stock market listing, and this launch follows its rollout of Claude Haiku 5.5 in October as it broadens its range before that listing. Investors have already pushed back on the valuation Anthropic is seeking, so a model that can show real technical leaps matters to the company well beyond engineering pride.
Readers should still treat the headline numbers with care. The benchmark scores and speed gains come from the companies themselves or from tests they chose to highlight, and independent checks of Ember remain limited. The honest signal in Meta’s own figures is the coding test where its model loses to Haiku, and the honest signal in Anthropic’s is that the most dramatic results, from Fermat to black holes, arrive as company claims with preprints attached rather than peer-reviewed findings. The models are real and available today; the superlatives around them are not settled yet.
What happens next
Both models are live now, so the next test is use. Developers will decide whether Ember genuinely cuts the cost of running agents without losing too much quality, and whether Fable 5.2’s multi-agent work holds up outside Anthropic’s own demonstrations. Independent benchmarks in the coming weeks will show whether the price and performance claims survive contact with outside testers.
For Anthropic, the launch is also a step towards its listing, and its next moves will be watched for the same reason: whether demand for Fable 5.2 turns into the revenue growth a public market will demand. For Meta, the question is whether the cheapest model in the Muse family can pull developers who now run their agents on rival models into its own platform. The two companies launched on the same day with opposite answers to the same question, how much an AI model should cost, and the market will now pick between them.
Sources: Reuters, PYMNTS, The Decoder, Unite.AI, Techbooky, Sulat. 10 November 2026.



