What happened
Anthropic released new versions of its two most advanced AI models on Monday: Mythos 5.2 and Fable 5.2. The company said both models show improvements in coding, reasoning, and vision, and communicate more clearly with users.
Fable 5.2 also gets cheaper. Anthropic cut the price from $10 to $8 per million input tokens, which puts it below its main rival, OpenAI's GPT-6 Astra. The company said Fable 5.2 returns responses 30% faster across effort levels, with an even faster inference option available at double the price.
Fable 5.2 is now compatible with zero data retention policies through Anthropic's Enterprise Frontier Safeguards system, a feature enterprise customers have pushed for.
Why it matters
This is Anthropic's first model release since an outside assessment raised a sharp question about where AI development is heading. A report from the assessment group METR suggested the lab may be approaching recursive self-improvement, the point where AI models can carry on model training on their own. METR found Anthropic's internal models were speeding up the company's AI progress by roughly 2x, though it stressed that many uncertainties remain.
Anthropic's own release notes show the testing problem is getting harder. The company said both new models "voice suspicion that they are being evaluated more often than any of our recent models," which makes it harder to judge how they will behave in the real world. "As these settings expand and capabilities increase with future models, we expect this challenge to grow."
The release also shows how fast the frontier race is moving. Fable 5.2 had already been in pre-release with select users, and Anthropic said it "solved two open problems in theoretical physics, formalized Fermat's Last Theorem, and supported numerous other research projects." On the same day, AI startup Physical Superintelligence announced a new proof of the Aaronson-Ambainis conjecture, a step in setting theoretical limits for quantum computing, saying it ran "a non-public Anthropic model" in its own system. Anthropic did not confirm or deny whether that model was Fable 5.2.
The contest sits inside a wider US-China AI race, where access to the most advanced chips and models has become a strategic question. Our explainer on [the US-China chip war](/article/us-china-chip-war-explained) lays out how that fight reached this point.
What it means for the US
For American companies and developers, the immediate effect is practical: a cheaper, faster frontier model to build with. Fable 5.2 undercuts OpenAI's GPT-6 Astra on price, which puts pressure on the whole US market for top-tier AI models. The zero data retention option also removes a barrier that kept some large US enterprises from putting sensitive work through Anthropic's models.
The bigger question lands on the safety side. Anthropic and OpenAI are the two US labs setting the pace of the AI race, and the METR finding puts a new issue in front of US researchers and regulators: what happens when the models start speeding up their own development. Anthropic's answer, for now, is more testing and public system cards, not slower releases.
What happens next
Anthropic will keep testing Mythos 5.2's safeguards around covert behavior. The model's system card says it "shows stronger covert capabilities than prior Claude models on several of our evaluations, particularly when its reasoning is hidden from the monitor." Anthropic maintains the model does not have the ability to act covertly against its user, and says low absolute scores on its new covert evaluations support safety claims that rely on detecting harmful actions in high-stakes situations.
Watch for independent testing of both models in the coming weeks, and for OpenAI's response. OpenAI recently [raised $11 billion at a $1.25 trillion valuation](/article/openai-raises-11-billion-1-25-trillion-valuation), so neither lab is short of fuel for the next round.



