Claude Fable 5.2 vs Mythos 5.2 vs Opus 5.5: Which to Use
The short answer
Fable 5.2 and Mythos 5.2 are one model with two safety settings. Pick Opus 5.5 by default, Fable 5.2 for the hardest long-running work, and Mythos 5.2 only if you do verified security or biology research. Anthropic released both on November 9, 2026:
- Claude Opus 5.5 — the default for most coding and knowledge tasks. $4 input / $20 output per million tokens.
- Claude Fable 5.2 — for ambitious, open-ended technical work: multi-day agent runs, large migrations, hard reasoning. $8 / $40, available to everyone.
- Claude Mythos 5.2 — the same model without most cyber and biology blocks, for organizations in Anthropic’s verification programs. $8 / $40.
The comparison
Prices from Anthropic’s pricing page and benchmark figures from Anthropic’s launch post, read November 10, 2026. Scores are each model’s highest available effort, with safeguards on unless noted.
| Fable 5.2 | Mythos 5.2 | Opus 5.5 | Fable 5.1 | |
|---|---|---|---|---|
| Model ID | claude-fable-5-2 | claude-mythos-5-2 | claude-opus-5-5 | claude-fable-5-1 |
| Input / output per MTok | $8 / $40 | $8 / $40 | $4 / $20 | $10 / $50 |
| Cache read | $0.20 | $0.20 | $0.20 | $0.25 |
| Batch | $4 / $20 | $4 / $20 | $2 / $10 | $5 / $25 |
| Fast mode | $16 / $80 | $16 / $80 (verified orgs) | $8 / $40 | — |
| Who can use it | All customers | Verified organizations | All customers | All customers |
| FrontierCode 2.0 | 49.1% | — | 45.5% | 32.4% |
| FrontierSWE v2 | 69.9% | — | 62.3% | 56.3% |
| OSWorld 2.1 (partial) | 89.3% | — | 87.2% | 83.5% |
| Humanity’s Last Exam (tools) | 74.2% | — | 67.7% | 65.6% |
| CVE Rediscovery | blocked | 89.6% | 81.3% (no safeguards) | — |
All three current models share a 1M-token context window and 128K max output. For other vendors’ prices see our current API prices.
What changed from Fable 5.1
- 20% cheaper. Input and output dropped from $10/$50 to $8/$40; cache reads from $0.25 to $0.20.
- Up to 30% faster output, per Anthropic, and fewer tokens per task at the default medium effort.
- Big jumps on agentic work. FrontierCode 2.0 rose from 32.4% to 49.1% and the Remote Labor Index from 17.9% to 41.3%. Fable 5.2 beats GPT-6 Astra (65.5%) and Gemini 4 Carbon (61.2%) on FrontierSWE v2 in Anthropic’s table, but Gemini 4 Carbon still leads AutomationBench (53.9% vs 51.4%).
- Fewer false blocks. Anthropic says Fable 5.2’s cyber and biology classifiers are as strict as Fable 5.1’s but refuse fewer harmless requests. When a request is blocked, Claude answers with Opus 4.8 (cyber) or Opus 5 (biology) instead.
- Breaking API changes. Edited conversation history now returns a 400 error on every account, and the old
computer_20251124tool is rejected outside Bedrock. Our Fable 5.2 migration guide covers each one.
When each model is the right pick
Opus 5.5 — the default. Anthropic’s launch post recommends starting here and moving up or down as needed. At half Fable’s price with 87–93% of its scores on most coding benchmarks, Opus 5.5 wins on cost for everyday pull requests, chat assistants and RAG.
Fable 5.2 — when Opus fails your evals. The gains are largest on long, open-ended work: Anthropic had a 1,000-agent team of Fable 5.2 formalize Fermat’s Last Theorem in 30 hours, where Opus 5.5 finished 26% in 18 hours. Fable 5.2 can also act as an advisor to cheaper models or an orchestrator that hands subtasks to Haiku or Sonnet, which keeps the bill down.
Mythos 5.2 — security and biology research only. Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 on its internal CVE Rediscovery test. Fable 5.2 blocks most penetration testing, exploit generation and binary vulnerability work, so a security team needs Mythos or Claude Security. See our explainer on Mythos trusted access for how the program works.
Limits to know
- No zero data retention. Fable 5.2 and Mythos 5.2 require 30-day data retention unless Anthropic expressly authorizes otherwise.
- Subscription caps. During the launch promotion, Fable counts toward up to 50% of weekly usage limits on Pro plans and standard Team seats; eligible plans also get one to three Fable usage resets.
- Benchmarks are vendor-reported. Treat the table as Anthropic’s numbers until independent evaluations land.
Last verified: November 10, 2026.