Anthropic Ships Claude Fable 5: First Mythos-Class Model Goes Public at 80.3% SWE-Bench Pro
Anthropic launched Claude Fable 5 on June 9, 2026 — the first publicly available model in its Mythos-class tier. For weeks, Mythos-class capabilities were restricted to cyberdefense partners in Project Glasswing. Fable 5 is the same underlying model made accessible to everyone, with classifiers that intercept fewer than 5% of sessions on average.
Key Benchmarks
| Benchmark | Fable 5 | Opus 4.8 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| SWE-Bench Pro | 80.3% | 69.2% | 58.6% | 54.2% |
| FrontierCode Diamond (xhigh) | 29.3% | 13.4% | 5.7% | — |
| GDPval-AA ELO | 1932 | 1890 | 1769 | 1314 |
| OSWorld-Verified (computer use) | 85.0%* | 83.4% | 78.7% | 76.2% |
| Terminal-Bench 2.1 | 88.0%* | 74.6% | 78.2% | 70.7% |
Starred scores are Mythos 5 (no safeguards); Fable 5 lands closer to Opus 4.8 on security-adjacent tasks.
SWE-Bench Pro uses freshly filed GitHub issues from actively maintained repos — no public ground-truth, no training data leakage. An 80.3% resolve rate means the model cleared four in five novel, multi-file engineering problems without human assistance. That is a 10-point improvement over Opus 4.8 and more than 20 over GPT-5.5.
The FrontierCode gap is harder to ignore. Cognition’s benchmark tests production-quality code under real engineering standards. Fable 5 scores 29.3%. Opus 4.8 scores 13.4%. GPT-5.5 scores 5.7%. The gaps compound as task complexity increases.
How the Safeguards Work
Fable 5 is the same model as Mythos 5, with classifiers layered on top. When a query touches cybersecurity, biology, chemistry, or distillation in ways the classifier flags as high-risk, the session falls back to Claude Opus 4.8 instead. Anthropic reported that in testing, 95% of Fable sessions ran entirely on Fable responses.
The classifier is tuned conservatively — some benign queries trigger it. Anthropic acknowledged this will be frustrating and said it will reduce false positives as safeguard quality improves.
For Glasswing partners and approved biology researchers, Claude Mythos 5 ships today with the classifiers lifted in those domains. All existing Mythos Preview users can upgrade.
Production Evidence
Stripe reported that Fable 5 completed a codebase-wide migration across 50 million lines of Ruby code in a single day. The same task would have taken a full engineering team more than two months manually. IMC, the proprietary trading group, said Fable 5 passed its trading analysis evaluations nearly across the board — factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis.
On Hebbia’s Finance Benchmark — testing reasoning at senior analyst level — Fable 5 posts the highest score of any model yet measured.
Pricing and Access
Both Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens — exactly double Opus 4.8 ($5/$25) and less than half the previous Mythos Preview price.
The rollout schedule for subscription plans:
- Through June 22: Fable 5 included at no extra cost on Pro, Max, Team, and seat-based Enterprise plans.
- June 23: Removed from subscription plans; usage credits required.
- Later: Restored to subscriptions once capacity allows.
The Claude API and consumption-based Enterprise plans have full access from today. API identifier: claude-fable-5.
What Changed in the Claude Lineup
Opus was previously Anthropic’s top commercial tier. It no longer is. The hierarchy now reads: Fable 5 for general users, Mythos 5 for restricted high-risk work, with Opus 4.8 as the intelligent fallback. The longer and more complex the task, the larger Fable 5’s advantage over Opus 4.8 — Anthropic’s own framing, consistent with the benchmark data.
Near-term: Anthropic said models of comparable capability will arrive in the coming months.