Claude Fable 5 is Anthropic's best model — state-of-the-art on nearly all benchmarks. It excels at long-horizon coding, vision tasks, and scientific research. Priced at $10/million input and $50/million output tokens. Available now via API and Claude Pro/Max/Enterprise plans.
Anthropic released Claude Fable 5 on June 9, 2026, and it immediately became one of the most capable AI models available. Unlike a typical incremental update, Fable 5 represents a genuine leap — compressing months of complex engineering into days, solving frontier physics problems with a fraction of the compute, and completing vision-only tasks that previous models couldn't handle at all. If you need maximum capability and can justify the cost, this is the model to use.
What Is Claude Fable 5?
Fable 5 is Anthropic's flagship model in the Claude Fable family — a tier above Opus 4.8 and designed for the most demanding use cases: massive codebase migrations, frontier scientific research, complex financial analysis, and long autonomous task sessions spanning millions of tokens. Anthropic describes it as 'state-of-the-art on nearly all tested benchmarks of AI capability.'
Claude Fable 5 Benchmarks and Performance
| Benchmark | Fable 5 Result | Why It Matters |
|---|---|---|
| FrontierBench (coding) | Highest score tested | Complex, multi-day coding tasks |
| Hebbia Finance Benchmark | Best tested | Deep financial analysis and reasoning |
| IMC Trading Analysis | Best tested | Real-world trading strategy evaluation |
| Frontier Physics Research | Best tested (1/3 the tokens of rivals) | Solves novel physics problems efficiently |
| OSWorld (computer use) | Top tier | Autonomous browser and desktop operation |
| Vision — Web Rebuild | Screenshot → full app | Reconstructs web apps from images alone |
Real-world results back the benchmarks. Stripe used Fable 5 to complete a 50-million-line Ruby migration in a single day — a task that would have taken two months manually. That is not an incremental improvement; it is a category shift in what AI-assisted engineering looks like.
Claude Fable 5 Pricing
| Token Type | Price per Million Tokens |
|---|---|
| Input tokens | $10 |
| Output tokens | $50 |
| Cached input (90% discount) | $1 |
Fable 5 is priced at approximately 50% less than Claude Mythos Preview — the restricted version available only to vetted cybersecurity and biology researchers. For enterprise teams, the prompt caching discount is meaningful: long-context tasks that repeatedly reference the same system prompt or codebase effectively cost $1/million input tokens after the first pass.
Key Capabilities
- Software engineering: Handles large migrations, complex implementations, and multi-day autonomous sessions without losing context
- Vision: Extracts precise data from scientific figures, rebuilds web apps from screenshots, completed Pokémon FireRed using vision-only input
- Long-context: Maintains focus across millions of tokens; memory integration tripled performance gains versus previous models
- Financial analysis: Top-ranked on both general finance benchmarks and trading-strategy evaluations
- Scientific research: Strongest model on frontier physics tasks, using one-third the reasoning tokens of competing models
- Agentic tasks: Plans and executes multi-step tasks autonomously with tool use, browser operation, and code execution
Fable 5 vs Claude Opus 4.8
| Factor | Fable 5 | Opus 4.8 |
|---|---|---|
| Raw capability | State-of-the-art across all benchmarks | Strong generalist, behind Fable 5 |
| Best use case | Frontier coding, research, long sessions | General enterprise tasks, agentic work |
| Input cost | $10/million tokens | Lower |
| Output cost | $50/million tokens | Lower |
| Availability | Pro, Max, Team, Enterprise, API | All plans including Free |
| Safety safeguards | Routes <5% of queries to Opus 4.8 | Standard guardrails |
Safety Features
Fable 5 launched with a three-layer safety framework that distinguishes it from the unrestricted Claude Mythos 5. Sensitive queries on cybersecurity attack planning, bioweapons, and model distillation are automatically routed to Claude Opus 4.8 via AI classifiers. This affects fewer than 5% of sessions on average and is invisible to legitimate users. Fable 5 complied with 0% of harmful single-message requests in red-team testing.
Who Should Use Claude Fable 5?
- Engineering teams running large-scale migrations or multi-day autonomous coding sessions
- Researchers in finance, physics, or life sciences who need frontier reasoning at scale
- Enterprises building agentic pipelines that require the highest accuracy on tool use and planning
- Developers using computer use to automate complex browser or desktop workflows
- Teams that hit the limits of Sonnet 5 or Opus 4.8 on precision-critical tasks
How to Access Claude Fable 5
Fable 5 is available immediately via the Claude API using the model ID `claude-fable-5`. It is also accessible in the Claude interface for Pro, Max, Team, and Enterprise subscribers. The model became available globally on July 1, 2026, after export controls were lifted on June 30. Free plan users do not have access to Fable 5.
Claude Fable 5 is the most capable general-purpose AI model available as of mid-2026. If your work hits the ceiling of Sonnet 5 or Opus 4.8 — especially in coding, long-context reasoning, or scientific tasks — Fable 5 is worth the premium. For typical day-to-day use, Sonnet 5 offers better cost-to-capability balance. But for anything that qualifies as frontier work, Fable 5 is the clear choice.