Claude Fable 5.2: Anthropic’s New Model Gets Faster and Cheaper
By Veer Solanki · · 945 words
Topics: AI, 1 Million Token Context Window, Adaptive Reasoning, Advanced AI Models, Advanced AI Reasoning, Advanced Coding Agents, Agentic AI, AI Agents
Anthropic released Claude Fable 5.2 on November 9, 2026. The model boasts increased capabilities, plus 30% faster outputs and a 20% price drop over Fable 5.1. It comes alongside Claude Mythos 5.2 (its private counterpart), less than two months after the 5.1 versions were released in September.
Fable is not a model you will use to edit an email but a model Anthropic wants you to use on complex tasks such as migrating a codebase, calculating a physics problem with no known solution, or a research loop that takes hours to run. Fable 5.2 makes the case that this tier is now cheap and fast enough to use regularly.
Same model?
Claude Fable and Claude Mythos are both based on the same model. Mythos is the less restricted one. Fable has safety measures to ensure the model does not cause issues in areas such as cybersecurity, biology and chemistry. It also has safety checks to ensure the model cannot be distilled.
Back in June, Anthropic released Fable 5 as their first generally available ‘Mythos-class’ model, which is a tier above Opus. However, Mythos had been restricted due to its high capabilities. It was instead only available to verified partners via Project Glasswing.
The safety measures are rarely hit. Anthropic claims less than 5% of sessions hit these restrictions. When they do, Claude will just fall back to a lower-tier model. For general technical tasks, Fable acts the same as Mythos. Mythos 5.2 is still restricted and only available to organisations within Anthropic’s Life Sciences and Cyber Verification Programs.
New features
Anthropic states that Claude Fable 5.2 and Claude Mythos 5.2 outperform Opus 5.5 across almost all capabilities. They also claim testers found the models easier to communicate with than previous Fable models. Key features include:
• Outputs are 30% faster.
• Pricing has dropped to $8 per million input tokens and $40 per million output tokens. Previously, it was $10 and $50.
• Math – 1000 Fable 5.2 agents formalised Fermat’s Last Theorem in 30 hours.
• Physics – Solved 2 open problems in black hole physics during testing.
• Robotics – Significant improvements in controlling physical robots, even on tasks it wasn’t trained for.
• Engineering – Helped Anthropic’s engineers complete migrations that were previously impossible. Claude now writes over 90% of the changes that are merged into Anthropic’s core codebase with human oversight.
The specifications have not changed from the previous generation. It has a 1M token context window, 128K output limit, always-on adaptive reasoning, and a June 2026 cutoff date.
Interesting stats
The most interesting stat from the start is how efficient Fable 5.2 is. On FrontierCode 2.0, Fable 5.2 at High effort roughly matched Opus 5.5 at Max effort, at about half the cost per task.
It also performed well in independent testing. It came first out of 36 models in the Vals Index, with a score of 71.31%. It came first out of 127 models in the Vibe Code Bench v1.1 and second out of 45 in the Terminal-Bench-Science with a score of 55.71%.
However, the average time to complete a full run of the Vals Index was around 44 minutes, at around $18 per test. This is a model that will take its time to think. As always, vendor benchmarks can be misleading, so it’s best to test it on your own workloads.
Should I use Fable or Opus?
Anthropic is clear that Fable is for power users. Dianne Penn from Anthropic told CNBC that Opus 5.5 is still the best choice for everyday tasks. However, Fable is for complex tasks in engineering, maths and robotics.
| Model | Input / output (per M tokens) | Default effort | Best for |
|---|---|---|---|
| Fable 5.2 | $8 / $40 | High | Hours-long agents, deep research, hard engineering |
| Opus 5.5 | $4 / $20 | Medium | Serious daily work |
| Sonnet 5.5 | $2 / $10 | High | Fast, capable general use |
| Fable 5.1 | $10 / $50 | High | Superseded |
Fable 5.2 is available on all paid Claude plans. If you are on an eligible plan,n you can get additional Fable usage for trial. Developers can access it by calling claude-fable-5-2 on the Claude Platform, Amazon Bedrock, Google Cloud and Microsoft Foundry. Some platforms offer a fast mode at a premium price.
Why now?
This is Anthropic’s first frontier release since CEO Dario Amodei called for slowing the pace of AI development, and it came after extended safety testing. Penn told Bloomberg that pacing development is not the same as pausing it. The Mythos 5.2 system card also states the less restricted model often felt it was being tested. As models improve, ve it becomes harder to test them.
Anthropic is also preparing for an IPO that could happen as soon as next week. Fable already made up 11% of spend on Anthropic models in the August AI Index from Ramp. Having a fast, cheap flagship model is ideal to release before going public.
The Fable series has had a rocky year. Fable 5 and Mythos 5 were suspended on June 12 due to US export controls. They were then reinstated on July 1 after the export controls were lifted. Anthropic has now released two point updates in around 2 months.
Overall
Fable 5.2 doesn’t feel like a huge step forward but addresses the main issue with Fable: le, its price. Getting the same performance as Opus 5.5 at Max effort, but at half the price per task, will make a difference to anyone using long agents. Most people should continue to use Opus 5.5. However, if you’re doing tasks that take hours, working with large codebases or tackling problems with no known solution, Fable 5.2 is now clearly the first port of call.