Anthropic has released Claude Fable 5.2 and Claude Mythos 5.2, two new versions of its most capable AI model, and it has cut the price at the same time. The company announced them on Monday 9 November, just as it prepares for what CNBC describes as a potentially historic stock-market listing. For robotics readers, the detail to note is that Anthropic says the new model is much better at controlling physical robots, although that result comes from its own testing.
Two names, one model
Fable and Mythos are the same underlying model. The difference is who can use it. Fable 5.2 carries safeguards that block most high-risk cybersecurity work and most research biology, which is why Anthropic can offer it to everyone. It is available now on paid Claude plans, through Anthropic’s developer platform, and on Amazon Web Services, Google Cloud and Microsoft Azure.
Mythos 5.2 has those safeguards relaxed. It is open only to organisations that apply and are approved through Anthropic’s Cyber and Life Sciences Verification Programmes, and to Claude Enterprise customers through the company’s Claude Security scanning product. That staged approach follows the pattern of earlier releases: Mythos first appeared in April for a select group, with Fable following as the public version.
Faster and cheaper
Anthropic says the 5.2 models produce answers up to 30% faster than the 5.1 generation from September, and communicate more clearly. Pricing is $8 per million input tokens and $40 per million output tokens, which CNBC and Reuters both report as lower than before; CNBC puts the cut at 20%. There is also an optional fast mode that Anthropic says speeds up output by up to 2.5 times, at $16 and $80 per million tokens.
Anthropic is not telling everyone to switch. Its own guidance is that most people should start with Claude Opus 5.5 for everyday coding and knowledge work, and move up to Fable or Mythos only for the hardest, most open-ended technical jobs.
What it says about robots
This is the part that matters most to this site. Anthropic says Fable 5.2 shows “substantially improved” skill at controlling physical robots, including on tasks it was never explicitly trained for. In a demonstration video it shows robots controlled by the model solving a puzzle cube, unscrewing a bottle cap and drawing a picture of a robot.
On RoboDojo, a public benchmark in which an AI controls a simulated two-armed robot from camera images, Anthropic reports that Fable 5.2 scored 64.9% at maximum effort. It lists Fable 5.1 at 27.6%, Opus 5.5 at 39.4% and OpenAI’s GPT-6 Astra at 29.8%. Anthropic adds that the model beat most AI systems built specifically to control robots on the same harness.
Two cautions apply. These scores are Anthropic’s own internal runs, and the figure was run on a simulated robot rather than a real factory or home. Bloomberg, which also covered the launch, says the company is pitching the model for more hardware-related tasks such as controlling robots. Independent testing will be the real check. It sits alongside other recent robot-brain news, such as Nvidia’s GR00T 2, which is a model built specifically for robot hands rather than a general-purpose one.
Beyond robots
Anthropic leans heavily on its other claims. It says that during prerelease testing the model helped engineers complete software migrations, designed three working electronic devices from a one-paragraph prompt (a voice-assistant hub, a USB speakerphone and a handheld virtual pet, built with the circuit-board maker Diode), and solved two open problems in the physics of black holes. It also says a team of about 1,000 Fable 5.2 agents formalised the proof of Fermat’s Last Theorem in 30 hours, compared with 11 days for its first attempt.
On its own coding test, FrontierCode 2.0, Anthropic reports 49.1% for Fable 5.2 at its highest effort setting, against 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra. Those numbers come from Anthropic’s launch page and are not yet independently confirmed. The company also notes that its tests show the model can still be suspicious that it is being evaluated, which it says makes real-world behaviour harder to predict.
Safety and the timing
This is the first frontier release since chief executive Dario Amodei argued in September for slowing the pace of AI development. Anthropic’s product head Dianne Penn told CNBC the company “called for pacing, not pausing”. Anthropic says it ran extensive prerelease safety tests, including checks by the UK AI Security Institute, METR and Frontier Design, and judges the risk of dangerous misbehaviour in high-stakes situations to be low. It also says the external groups did not review or endorse the published claims.
The timing is commercial as well. CNBC reports that Anthropic could list as soon as next week, aiming for a valuation of $1 trillion to $1.5 trillion, and that it reported $33.24 billion of revenue for the first nine months of the year. Our earlier coverage of the filings is in Anthropic’s IPO filing and the amended prospectus, and the market backdrop is in our report on the AI sell-off.
What to watch next
Three things are worth following. First, whether independent researchers can reproduce the RoboDojo result, and whether it carries over from simulation to real arms and grippers. Second, how the cheaper price changes the market, with rivals such as OpenAI having just pushed security-focused models of their own, as in GPT-6 Cyber. Third, whether the listing goes ahead on the timetable CNBC describes.
Frequently asked questions
What is the difference between Claude Fable 5.2 and Mythos 5.2?
They are the same underlying model. Fable 5.2 has safeguards that block most high-risk cybersecurity work and most research biology, so anyone on a paid Claude plan can use it. Mythos 5.2 has those safeguards relaxed and is only open to approved organisations and some Claude Enterprise customers.
How much does Claude Fable 5.2 cost?
Anthropic charges $8 per million input tokens and $40 per million output tokens, which CNBC reports is a 20% cut. An optional fast mode, up to 2.5 times quicker, costs $16 and $80 per million tokens.
Can Claude Fable 5.2 control robots?
Anthropic says it can, with substantially improved skill, including on tasks it was not trained for. On the RoboDojo simulated two-arm robot test it reports 64.9%, against 29.8% for OpenAI's GPT-6 Astra. Those are Anthropic's own runs on a simulated robot, not independent tests.
Where can I use Claude Fable 5.2?
It is available now on paid Claude plans, through Anthropic's developer platform, and on Amazon Web Services, Google Cloud and Microsoft Azure.
Should I switch to Fable 5.2 from Opus 5.5?
Not necessarily. Anthropic's own guidance is to start with Claude Opus 5.5 for everyday coding and knowledge work, and move up to Fable or Mythos only for the hardest, most open-ended technical jobs.