Claude Fable 5.2 and Mythos 5.2: pricing, benchmarks, access
Anthropic upgraded its Mythos-tier models on November 9, 2026: Claude Fable 5.2 and Claude Mythos 5.2 run up to 30% faster than the 5.1 pair, cost less, and beat Opus 5.5 on nearly every evaluation Anthropic published.
- Price cut and new fast mode — Fable 5.2 and Mythos 5.2 list at $8 per million input tokens and $40 per million output, with an optional fast mode at $16/$80 that speeds token generation by up to 2.5x.
- Same model, two doors — Fable and Mythos are the same underlying model: Fable carries cyber and biology safeguards and is open to all users, while Mythos requires approval through a verification program.
- Ahead of Opus 5.5 on most evals — Anthropic's own table puts Fable 5.2 at 49.1% on FrontierCode 2.0 and 69.9% on SWE v2, against 45.5% and 62.3% for Opus 5.5.
- 1M context, 128K max output — The Claude Platform model docs list a 1M-token context window, 128K max output tokens and a June 2026 knowledge cutoff for both 5.2 models.
What changed on November 9
Related guide: Claude Agent Memory: Anthropic's Dreaming Fixes Stale Notes
Three things moved at once, and they are worth separating because they carry different risk.
- Price down. Anthropic lists Fable 5.2 and Mythos 5.2 at $8 per million input tokens and $40 per million output, with cached input at $1.25. Fable 5.1 and Mythos 5.1 were $10/$50, so the flagship tier got cheaper while getting stronger.
- Speed up. Anthropic says the 5.2 models "generate outputs up to 30% faster than Claude Fable 5.1 and Claude Mythos 5.1".
- A paid fast mode. An optional fast mode costs $16 per million input and $80 per million output and speeds up token generation by up to 2.5 times. It is available on Claude Code, the Claude Platform, Claude Models on Google Cloud and Microsoft Foundry — with Mythos fast mode limited to organizations in the verification programs.
Availability is broad. Fable 5.2 is on all platforms — the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure — with the API model ID claude-fable-5-2. Subscribers on eligible paid plans also get extra Fable usage for launch week: Fable counts for up to 50% of weekly usage limits on Pro plans and standard Team seats, and plans that already include Fable get one to three resets based on subscription length.
The model docs are blunt about which one to reach for: start with Claude Opus 5.5 for most workloads, and move to Claude Fable 5.2 "for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short".
The benchmark table, read properly
Related guide: Claude vs GPT in 2026: which one should you actually pay for?
Anthropic published a comparison grid with Fable 5.2, Fable 5.1, Opus 5.5, GPT-6 Astra and Gemini 4 Carbon. Selected rows, all at each model's highest available effort:
| Evaluation | Fable 5.2 | Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| FrontierCode 2.0 (agentic coding) | 49.1% | 45.5% | 42.8% |
| SWE v2 (agentic coding) | 69.9% | 62.3% | 65.5% |
| WANDR (agentic search, relaxed) | 82.1% | 72.3% | — |
| OSWorld 2.1 (computer use) | 89.3% | 87.2% | 79.5% |
| Humanity's Last Exam (with tools) | 74.2% | 67.7% | 57.2% |
| Chartography (chart reading) | 73.4% | 64.4% | 71.0% |
| RoboDojo Sim (robotics) | 64.9% | 39.4% | 29.8% |
The interesting part is not the leaderboard, it is the cost-per-task curves Anthropic published alongside it. On FrontierCode 2.0, Fable 5.2 at High effort scores about the same as Opus 5.5 at Max effort (45.3% versus 45.5%) "for roughly half the cost per task". On WANDR, Fable 5.2 at Medium effort matches Opus 5.5's top score "for about 40% of the cost per task". On Humanity's Last Exam, Fable 5.2 at Low effort matches Opus 5.5's best score "for less than one-tenth of the cost per task".
That is the real argument for the premium tier, and it only works if you actually tune the effort setting. The default effort differs by model — high for Fable 5.2 and Sonnet 5.5, medium for Opus 5.5 and Haiku 5.5 — so a naive migration can quietly bill you more than the one you left.
Fable or Mythos: same model, different doors
Related guide: Claude 5.1 (Fable) in 2026: Anthropic's Mythos flagship — and why Opus 5 steals it
Fable and Mythos are the same underlying model. The difference is safeguards. Fable carries strong cybersecurity and biology safeguards that make it safe to offer to every user; Anthropic says those safeguards are "as robust as those of Fable 5.1 but produce fewer unnecessary blocks". Mythos runs with those restrictions lifted and is gated behind application.
Access routes, as published:
- Fable 5.2 — open to all users on all platforms, including AWS, Google Cloud and Azure.
- Mythos 5.2 — application through the Cyber Verification Program for cybersecurity work, or the Life Sciences Verification Program for biomedical research.
- Claude Enterprise — can use Claude Security, which scans codebases for vulnerabilities using Anthropic's most capable models, now including Mythos 5.2.
On cyber specifically, Anthropic reports that on its internal CVE Rediscovery evaluation — a run against 1,829 published CVEs in October 2026 — Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1, and that one tester found earlier models overstated the severity of roughly 1 in 20 findings while Mythos 5.2 overstated none. The same caveat applies to every number here: these are Anthropic's own evaluations, with each lab using unique tasks, and none of the outside companies quoted in the post reviewed the claims.
What Anthropic says it shipped into production
Related guide: Claude Marketplace and the plugin directory: what changed for MCP builders this week
Beyond benchmarks, the launch post carries several engineering claims that are more useful than the scores for anyone running agents in production.
- Over 90% of the changes merged into Anthropic's core codebase are now written by Claude, with human guidance.
- A team of agents running an earlier Fable 5.2 build migrated a critical production service from Python to Rust; the service now "starts about 26 times faster".
- As an experiment, one engineer had Fable 5.2 design a prototype operating system from an empty folder. The model coordinated over 1,000 subagents for eight days and produced a working OS with a core, disk, keyboard and network drivers, permanent storage, a login screen and an installer.
- During prerelease testing, Anthropic says the models "formalized the proof of Fermat's Last Theorem in 30 hours" and solved two open problems in theoretical physics concerning black holes.
There is also a new safety surface worth knowing about if you handle regulated data. Preserved thinking now applies to all accounts, preventing distillation attempts by blocking edits to Claude's prior reasoning context. Enterprise Frontier Safeguards — automated monitoring that watches usage patterns across requests rather than single responses — is in private beta with customers including Bridgewater, Salesforce and Stripe, and runs on prompts and responses stored in customer infrastructure so it stays compatible with zero data retention. Eligible customers can keep zero data retention before implementing it; for everyone else Anthropic applies 30-day retention for safety monitoring.
Who should actually switch
Related guide: Claude Haiku 5.5: Anthropic Cuts Its Small Model's Price by 90%
Use Fable 5.2 when you can name the workload: long-horizon agent runs that exceed Opus 5.5 at higher effort, large migrations where a failed attempt costs more than the token bill, or research-grade reasoning where a 70% versus 68% swing decides the outcome. Anthropic also gives Fable 5.2 two orchestration roles that are worth knowing about — Advisor, where Haiku, Sonnet or Opus models consult it on harder problems, and Orchestrator, where it delegates simpler tasks to smaller models. That second pattern is often the cheapest way to use the flagship tier: a strong model planning, cheap models executing.
Stay on Opus 5.5 for everything else. At $4/$20 it is the value tier Anthropic itself points at first, and on the published cost curves it lands within a few points of Fable 5.2 on most real workloads. If you want the cheap end, Sonnet 5.5 is $2/$10 and Haiku 5.5 starts at $0.10/$0.50 — both with the same 1M-token context window and 128K max output.
Stay on Opus 5.5 otherwise. At half the flagship's price with a 1M-token context window, it remains the correct default, and Fable 5.2's new fast mode at $16/$80 only makes sense when wall-clock latency is the binding constraint.
How to check in one afternoon
Related guide: The UK can test frontier AI models again: what the 17-day freeze and the U-turn actually changed
Take the three tasks where Opus 5.5 currently fails or stalls. Run each on Opus 5.5 at Max effort and on Fable 5.2 at Low, Medium and High effort. Record the tokens spent, not just the score — Anthropic's own charts are cost-per-task, and the flagship only wins if the quality delta survives the token delta at your volumes. Then test one thing that matters operationally: whether the cheaper model can be instructed to escalate to Fable 5.2 when a decision crosses a line, instead of calling the flagship on every turn.
One last piece of housekeeping: the model docs commit Fable 5.2 to retirement "not sooner than November 9, 2027", and Opus 5.5 to "not sooner than September 22, 2027". If you build on the Mythos tier today, you have a year of runway — but pin model IDs rather than aliases if you want that promise to mean anything at 3am.
Frequently asked questions
Price cut and new fast mode
Fable 5.2 and Mythos 5.2 list at $8 per million input tokens and $40 per million output, with an optional fast mode at $16/$80 that speeds token generation by up to 2.5x.
Same model, two doors
Fable and Mythos are the same underlying model: Fable carries cyber and biology safeguards and is open to all users, while Mythos requires approval through a verification program.
Ahead of Opus 5.5 on most evals
Anthropic's own table puts Fable 5.2 at 49.1% on FrontierCode 2.0 and 69.9% on SWE v2, against 45.5% and 62.3% for Opus 5.5.
1M context, 128K max output
The Claude Platform model docs list a 1M-token context window, 128K max output tokens and a June 2026 knowledge cutoff for both 5.2 models.
Prices and benchmark figures are standard published rates as of November 10, 2026 and can change without notice. Some outbound links may be affiliate links — see our disclosure. Found an error? Tell us and we'll correct it in public.