Anthropic released Claude Fable 5.2 on November 9, 2026. The flagship model is said to respond up to 30 percent faster and costs about 20 percent less than Fable 5.1. On ChatX, Fable 5.2 replaces the previous model. We ran both versions through the same tasks and worked out how many tokens a response costs now.
What Anthropic is promising
Fable is Anthropic’s line for especially hard tasks: long analyses, demanding code, multi-step reasoning. According to Anthropic, version 5.2 improves most in coding, science tasks and reading charts. Testers reportedly got responses up to 30 percent faster than with Fable 5.1. On top of that, input and output prices drop by a fifth.
One experiment by Anthropic shows just how capable the model is. A cluster of 1,000 Fable instances fully formalized the proof of Fermat’s Last Theorem in 30 hours, meaning it wrote the proof out step by step in a form a machine can verify. An earlier, unreleased model needed eleven days for the same job.
Fable thinks before every response. There is no mode without thinking. On ChatX, you control how thoroughly it reasons with the Low, Medium and High levels. ChatX doesn’t display the thinking steps themselves. They only count toward the token bill.
The key benchmarks
The numbers come from Anthropic. For comparison, we’ve added Claude Opus 5.5, which is also available on ChatX.
| Test | Fable 5.2 | Fable 5.1 | Opus 5.5 |
|---|---|---|---|
| Coding (FrontierSWE v2) | 69.9% | 56.3% | 62.3% |
| Science tasks (Terminal-Bench-Science) | 74.6% | 52.6% | 58.7% |
| Expert-level questions (Humanity’s Last Exam, with tools) | 74.2% | 65.6% | 67.7% |
| Reading charts (Chartography) | 73.4% | 44.8% | 64.4% |
The biggest jump is in charts. If you upload an image with line graphs or bar charts, Fable 5.2 reads the values far more reliably than its predecessor.
Our test: 5.1 vs. 5.2
On November 10, we gave both models three everyday tasks: a math word problem about two trains, a short email to a landlord and a PHP function that validates an IBAN. The times are averages of three runs at the Medium level. The coding task ran once at High.
| Task | Fable 5.1 | Fable 5.2 |
|---|---|---|
| Math problem | 10.7 seconds, about 11,100 tokens | 6.8 seconds, about 5,100 tokens |
| Email to the landlord | 5.4 seconds, about 4,100 tokens | 4.4 seconds, about 3,600 tokens |
| IBAN validation in PHP | 8.4 seconds, about 8,000 tokens | 13.7 seconds, about 13,000 tokens |
On the math problem and the email, Fable 5.2 was clearly faster. For the math problem, it reached the same correct answer with just under half the tokens. On the code, though, it thought longer and wrote a somewhat more thorough check. Both functions were correct.
The gap was widest at the Low level. Fable 5.1 made a mistake on the math problem and came up with 10:04. Fable 5.2 still got it right there, 10:11, and even double-checked its answer.
What a response costs on ChatX
ChatX bills based on what the provider actually charges. Since Fable 5.2 is cheaper, the price drops automatically. A typical exchange with a short question and a response of about 500 words now costs around 7,300 tokens instead of around 9,100 tokens before. Thinking counts toward that. At High, Fable almost always thinks at length, at Low much less.
You can use Fable 5.2 on ChatX with a free account. Tokens come out of your daily allowance, your subscription or token packs you’ve bought.
Which thinking level is worth it
- Low: for writing, emails and summaries. In our test, 5.2 was reliable on a math problem even at this level.
- Medium: for analyses, comparisons and step-by-step calculations. This is where the speed advantage over 5.1 was biggest.
- High: for tricky code and long documents. Expect longer responses and more tokens.
For everyday questions, Claude Sonnet 5 or Opus 5.5 are usually enough, and they cost less. Fable 5.2 really shines when a task is genuinely hard or when you need a chart analyzed.