Anthropic released Claude Fable 5.2 on November 9, 2026, and two numbers from its announcement are worth separating. The first is the headline score: 49.1% on FrontierCode 2.0, an agentic coding test, which Anthropic lists above Opus 5.5 at 45.5%, GPT-6 Astra at 42.8% and Gemini 4 Carbon at 41.1%. The second is a quieter line: at High effort, Fable 5.2 scores 45.3% on the same test, close to Opus 5.5's best result, at about half the cost per task. The second number is the one most developers should care about. (Sources: Anthropic, VKTR)
Quick summary: Fable 5.2 costs $8 per million input tokens and $40 per million output tokens, down from $10 and $50 for Fable 5.1, per Anthropic's models overview. Anthropic says it is up to 30% faster than Fable 5.1 and scores 49.1% on FrontierCode 2.0 at its Xhigh setting. Every figure is from Anthropic's own testing. Anthropic itself still recommends Opus 5.5 as the starting point for most coding and knowledge work, and treats Fable as the step up.
What launched on November 9
There are two models on one underlying system. Claude Fable 5.2 is available to all users, with safeguards around cybersecurity and biology. Claude Mythos 5.2 is the same underlying model without those safeguards, limited to vetted organizations in cybersecurity and life sciences through verification programs. Anthropic says Fable 5.2 is on all platforms including Amazon Web Services, Google Cloud and Microsoft Azure, and the API model ID is claude-fable-5-2. Anthropic also amended the post on November 11 to clarify one sentence and a footnote in its mathematical formalization section. (Source: Anthropic)
- Standard pricing: $8 per million input tokens and $40 per million output tokens, per Anthropic.
- Fast mode (optional): $16 and $80, with generation up to 2.5x faster, available in Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry.
- Speed: output up to 30% faster than Fable 5.1 and Mythos 5.1, an Anthropic claim.
- Previous price: Anthropic's models overview lists Fable 5.1 at $10 input and $50 output, so the cut is 20%.
- Consumer plans: Anthropic says eligible paid plans get one to three Fable resets based on subscription length, and Fable is included for up to 50% of weekly usage limits on Pro and standard Team seats.
- Retention: Anthropic says an Enterprise Frontier Safeguards private beta (Bridgewater, Salesforce and Stripe) is compatible with zero data retention, while other customers have 30-day retention for safety monitoring. TechEnclave reports the same 30-day requirement on the Claude Platform unless Anthropic has authorized zero retention.
The benchmark table
Anthropic's table compares Fable 5.2 with Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Scores generally use each model's highest available effort, with exceptions Anthropic marks, so check the footnotes. For example, Fable 5.2's FrontierCode figure is at Xhigh, while Gemini 4 Carbon's is at Medium. Here is the coding and agent slice. (Source: Anthropic)
| Benchmark | Fable 5.2 | Fable 5.1 | Opus 5.5 | GPT-6 Astra | Gemini 4 Carbon |
|---|---|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% (Xhigh) | 32.4% (Xhigh) | 45.5% | 42.8% | 41.1% (Medium) |
| FrontierSWE v2 (agentic coding) | 69.9% | 56.3% | 62.3% | 65.5% | 61.2% |
| OSWorld 2.1 (computer use) | 89.3% | 83.5% | 87.2% | 79.5% | 81.5% |
| AutomationBench (business workflows) | 51.4% | 31.4% | 42.5% | 41.4% | 53.9% |
| Humanity's Last Exam, with tools | 74.2% | 65.6% | 67.7% | 57.2% | not reported |
Two things stand out. Fable 5.2 improves on Fable 5.1 by a wide margin on coding, 49.1% against 32.4% on FrontierCode 2.0. And it does not win everything: Gemini 4 Carbon edges it on AutomationBench, 53.9% to 51.4%, in Anthropic's own table. All of these are Anthropic-run, and TechEnclave's own source list for the launch cites Anthropic's announcement and docs, with 9to5Mac as the only independent reporting. We found no independent hands-on test with measured results as of November 12.
The cost-per-task claim, unpacked
Per-token price is a poor guide to what a job costs, because a model that thinks less can finish cheaper even at a higher rate. Anthropic's page makes three effort-level claims worth knowing. At High effort, Fable 5.2 scores 45.3% on FrontierCode 2.0, near Opus 5.5's best of 45.5%, at about half the cost per task. At Medium effort, it matches Opus 5.5's top score on the WANDR agentic search test for about 40% of the cost per task. At Low effort, it matches Opus 5.5's best Humanity's Last Exam score for less than one-tenth of the cost per task. (Source: Anthropic)
Here is our inference, not Anthropic's: VKTR reports Opus 5.5's input price at $4 per million tokens, so Fable 5.2's $8 is about twice as much per input token. For Fable to cost half as much per task at similar accuracy, it has to use far fewer tokens on those tasks, or Opus 5.5 at its top setting has to spend a lot on reasoning. We did not see token counts in the sources we read. That is why the right reading is "promising on Anthropic's tasks" and not "cheaper everywhere."
When you compare models, hold effort constant and measure cost per completed task, not cost per token. Two models at the same per-token price can differ several-fold on a real job.
Real-world claims, taken with a grain of salt
Anthropic's page leans on dramatic examples. It says over 90% of changes merged into its core codebase are written by Claude with human guidance. A team of agents using an earlier Fable 5.2 version migrated a production service from Python to Rust, and Anthropic says the service starts about 26 times faster. An experiment coordinating over 1,000 subagents for eight days produced a prototype operating system from an empty folder. These are Anthropic's descriptions of selected projects, they do not tell you the failure rate, and a startup-time speedup on one internal service is not a standardized benchmark. TechEnclave makes the same point about the Rust figure. (Sources: Anthropic, TechEnclave)
Customer quotes on the page are also worth reading carefully. Vercel's Joe Haddad is quoted saying, "I haven't had to say 'No, I meant...' a single time," and Cognition's Adhyyan Sekhsaria says the model "says it doesn't know things instead of hallucinating." They are the kind of praise that fits a launch page, not controlled evidence.
Safeguards and why they matter for developers
Developers who tripped refusals on earlier Fable models will care about one number. Anthropic says Fable 5.2 has about 55% fewer safety-classifier blocks than Fable 5.1: about 60% fewer in cybersecurity and 40% fewer in biology. Blocked requests fall back to Opus 4.8 for cybersecurity and Opus 5 for biology. Anthropic also reports the models voice suspicion of being evaluated more often than recent models, which it says complicates evaluation. For day-to-day coding that means fewer false stops, but also a reminder that benchmark behavior may differ from production behavior. (Source: Anthropic)
Which Claude tier for which job, per Anthropic's guidance
| Situation | Start with | Why |
|---|---|---|
| Most coding and knowledge tasks | Opus 5.5 | Anthropic recommends it as the starting point, then moving up or down as needed. |
| Hard agentic jobs where retries cost real time | Fable 5.2 at High effort | Anthropic's best coding score at 49.1% on Xhigh, and 45.3% at High for about half Opus 5.5's per-task cost on its tests. |
| Cheap, high-volume work | Sonnet 5.5 or Haiku 5.5 | Anthropic names them as the step down from Opus. |
| Latency-sensitive loops | Fable 5.2 fast mode | Up to 2.5x faster at double the standard rates ($16 and $80). |
01How much does Claude Fable 5.2 cost?
Anthropic lists $8 per million input tokens and $40 per million output tokens. Fast mode is $16 and $80. Fable 5.1 was listed at $10 and $50.
02Is Fable 5.2 better than Opus 5.5 at coding?
On Anthropic's FrontierCode 2.0 test, 49.1% at Xhigh against 45.5%. At High effort it scores 45.3%, roughly level. These are vendor-run results, and Anthropic still recommends starting with Opus 5.5.
03What is the difference between Fable 5.2 and Mythos 5.2?
Same underlying model. Fable 5.2 is available to all users with cybersecurity and biology safeguards. Mythos 5.2 has no such safeguards and is limited to organizations in Anthropic's verification programs.
04Is there independent testing yet?
Not that we found as of November 12. The numbers here come from Anthropic, and outlets covering the launch largely relay its figures.
05Does it work in Claude Code?
Anthropic lists fast mode as available in Claude Code, and says Fable is included for eligible paid plans up to 50% of weekly usage limits on Pro and standard Team seats.
How to evaluate it without burning budget
Pick five tasks you already know the answer to, such as a bug you fixed last month, a refactor, a test-writing job, a code review and a small feature. Run them at Medium and High effort. Write down cost, wall-clock time and whether the output was mergeable. If Fable at Medium handles your hard cases that a cheaper tier fails, you have found its place in your stack. If the cheaper tier does fine, you saved money by checking. That is a better use of an afternoon than reading another benchmark chart.
If you would rather not manage several subscriptions just to run that kind of comparison, LumiChats gives you one place to try different AI models on the same prompt and see which answer you actually prefer. It keeps the decision grounded in your own work instead of a launch-day chart.
