Claude Fable 5: First Public Mythos-Class Model — Pricing, Benchmarks & Developer Guide
Anthropic's Claude Fable 5 is the first public Mythos-class AI model at $10/M input and $50/M output. Full pricing breakdown, benchmark analysis, and developer guide.
2026年6月11日 · 阅读约 5 分钟
Core Conclusion
Anthropic launched Claude Fable 5 on June 9, 2026 — the first public Mythos-class AI model, priced at $10/M input tokens and $50/M output tokens. It's the most capable model Anthropic has ever made generally available, but comes with novel safety guardrails that silently reroute sensitive queries to Claude Opus 4.8. For developers, this means access to frontier-level coding ability with caveats: priced at 2x Opus 4.8, with ~5% of sessions hitting a safety fallback. Early adopters including Stripe, Cursor, and GitHub report dramatically improved long-horizon task performance.
What Is Claude Fable 5?
Claude Fable 5 is the public-facing version of Anthropic's Mythos-class model — a tier previously restricted to government cybersecurity partners through Project Glasswing. The underlying model is the same as Claude Mythos 5, but Fable 5 adds safety classifiers that detect and reroute high-risk queries (cybersecurity, biology, distillation) to Opus 4.8.
Key specs:
| Spec | Value |
|---|---|
| Release date | June 9, 2026 |
| Input price | $10 / million tokens |
| Output price | $50 / million tokens |
| Output context | 128K tokens |
| Model ID | claude-fable-5 |
| Availability | API + subscription (staged rollout) |
| Safety model | Classifier-based fallback to Opus 4.8 (~5% of sessions) |
Source: Anthropic Blog — Claude Fable 5 and Claude Mythos 5
Pricing Deep Dive
At $10/M input and $50/M output, Fable 5 lists at 2x the price of Claude Opus 4.8 ($5/$25) and 2x GPT-5.5 on input ($5/$20). Compared to the previous Mythos Preview pricing (which was undisclosed but described as "substantially more"), it's less than half the cost.
Cost comparison for common workloads:
| Workload | Fable 5 | Opus 4.8 | GPT-5.5 |
|---|---|---|---|
| Code review (10K tokens in, 2K out) | $0.20 | $0.10 | $0.09 |
| Full PR analysis (50K in, 10K out) | $1.00 | $0.50 | $0.45 |
| Multi-file refactor (200K in, 50K out) | $4.50 | $2.25 | $2.00 |
| Long-horizon agent session (1M in, 200K out) | $20.00 | $10.00 | $9.00 |
The premium is substantial, but Stripe's reported result — compressing "months of engineering into days" on a 50-million-line Ruby codebase — suggests the ROI is there for complex tasks.
Source: Finout.io pricing analysis, Anthropic official pricing page
What Makes Fable 5 Different?
1. Software Engineering Performance
Fable 5 scored highest among frontier models on Cognition's FrontierCode evaluation, which tests real-world production code quality. Stripe documented a codebase-wide migration that would have taken "a whole team over two months" — completed in a single day by Fable 5.
Cursor CEO Michael Truell stated: "Claude Fable 5 is the state of the art model on CursorBench. It's opened up a class of long-horizon problems that were out of reach for earlier models."
GitHub CPO Mario Rodriguez added: "In our early testing, it took on complex, long-horizon coding tasks with a level of autonomy and reliability that exceeded previous benchmarks."
2. Knowledge Work and Finance
On Hebbia's Finance Benchmark for senior-level reasoning, Fable 5 achieved the highest score, with substantial gains in document-based reasoning, chart interpretation, and problem solving. Trading firm IMC reported Fable 5 "aced their trading-analysis evaluations nearly across the board."
3. Vision and Gaming
Fable 5 can rebuild a web app's source code from screenshots alone, extract precise numbers from scientific figures, and even beat Pokémon FireRed with a minimal vision-only harness — something previous Claude models struggled with even using complex tooling.
4. Memory and Autonomous Work
Fable 5 stayed focused across millions of tokens during long-running tasks. When playing Slay the Spire with persistent file-based memory, it reached the game's final act three times more often than Opus 4.8.
The Safety Reroute: What Developers Need to Know
Fable 5's most controversial feature is its safety system: when classifiers detect cybersecurity, biology/chemistry, or distillation-related queries, the response is silently handled by Opus 4.8 instead. Users are informed when this happens.
Developer impact:
- ~5% of sessions trigger a fallback
- Jailbreak resistance is the strongest of any tested model (zero harmful responses across 30 different jailbreak techniques)
- External red-teaming (1,000+ hours of bug bounty) found no universal jailbreaks yet
- UK AISI made "progress towards one within a brief initial testing window"
For most developers doing standard coding work, the fallback rate is minimal. But developers working on security tooling, penetration testing, or research involving exploit development will hit these guardrails.
💰 How Developers Earn Money With Claude Fable 5 Frontiers of AI capability like Fable 5 unlock new earning opportunities — from building production-grade AI agents for enterprise clients to creating specialized tooling that leverages long-horizon reasoning. The model's ability to complete codebase-wide migrations autonomously means developers can offer "months of engineering in days" as a premium service.
Availability and Rollout
Fable 5 is available immediately via the Claude API (claude-fable-5) and consumption-based Enterprise plans. For subscription plans, Anthropic is rolling out access in stages:
- Today: Pro and Team plan users get limited access
- Next 1-2 weeks: Gradual expansion to all subscribers
- Enterprise: Fully available from day one on consumption plans
Claude Mythos 5 (same model, safeguards lifted for cyber defense) remains restricted to Project Glasswing partners, with a biology trusted access program coming soon.
参考视频 / Further Reading
- Anthropic Blog: Claude Fable 5 and Claude Mythos 5 Announcement
- Finout.io: Fable 5 vs Mythos 5 Pricing Analysis
- Codersera: Fable 5 Launch Guide
Next Steps / 相关阅读
- 📚 Learn Claude Code → Claude Code Tutorials
- 💰 See how developers earn with Anthropic tools → Claude Earnings Cases
Claude Fable 5, Anthropic, Mythos, Opus 4.8, AI coding model, frontier AI, agentic coding, software engineering AI, AI pricing 2026