Claude Fable 5 is Anthropic’s most powerful AI model available to the public, launched on June 9, 2026. It belongs to a new tier called Mythos-class — sitting above the entire Opus lineup — and it is the biggest capability leap Anthropic has released in years. If you’ve been tracking AI tools, this one is worth understanding in detail.
What Is Claude Fable 5?
Claude Fable 5 is the publicly accessible version of Anthropic’s top-secret Mythos model. Anthropic originally launched Mythos in April 2026 under a restricted program called Project Glasswing, limiting access to a small number of cybersecurity and critical infrastructure partners because of the model’s advanced ability to identify software vulnerabilities.
Fable 5 changes that. It is the same underlying model as Claude Mythos 5 — sharing identical weights — but with safety classifiers enabled so it can be released to the general public. Those classifiers block responses in high-risk areas like cybersecurity, biology, chemistry, and distillation, automatically routing those queries to Claude Opus 4.8 instead. According to Anthropic, these classifiers trigger in fewer than 5% of sessions on average.
Claude Fable 5 Key Specs
- Model ID: claude-fable-5
- Context window: 1 million tokens
- Max output: 128,000 tokens
- Launch date: June 9, 2026
- Pricing: $10 per million input tokens / $50 per million output tokens
- Data retention: 30 days (mandatory; zero data retention not available)
- Availability: Claude API, Amazon Bedrock, Vertex AI, Microsoft Foundry, GitHub Copilot
Benchmark Performance: How Good Is It?
Anthropic calls Fable 5 its strongest publicly available model ever — and the benchmarks back that up. On SWE-Bench Pro, a rigorous test for solving real software engineering issues from public GitHub repositories without any assistance, Fable 5 scores 80.3%. For comparison, Claude Opus 4.8 scores 69.2%, GPT-5.5 lands at 58.6%, and Gemini 3.1 Pro sits at 54.2%. That is an 11-point lead over Anthropic’s own previous best, and roughly 22 points ahead of GPT-5.5.
On SWE-bench Verified, Fable 5 reaches 95%, compared to 88.6% for Opus 4.8. The performance gap is most pronounced on long, complex, multi-step tasks — exactly the kind that matter most in real production environments.
Developer and AI researcher Andrej Karpathy described it as “a major-version-bump-deserving step change forward.”
Real-World Performance: The Stripe Example
One of the most striking real-world demonstrations came from Stripe, which tested Fable 5 on a 50-million-line Ruby codebase. The model completed a full codebase-wide migration in a single day — work that would have taken a human engineering team more than two months. That is not a benchmark. That is production-grade performance on a system most AI tools would struggle to even load into memory.
The 1 million token context window is a big reason this is possible. You can load entire codebases, documentation sets, and conversation histories simultaneously, while still getting up to 128,000 tokens of output per response.
Claude Fable 5 vs Claude Mythos 5: What’s the Difference?
This is the question most people are asking. The short answer: they are the same model. The practical difference is access and safety classifiers.
- Claude Fable 5 — Full safety classifiers active. Available to everyone via the API and subscription plans. High-risk queries are rerouted to Opus 4.8.
- Claude Mythos 5 — Safety classifiers lifted in specific areas like cybersecurity. Restricted to vetted Project Glasswing partners and not publicly available.
For the vast majority of use cases — coding, writing, research, analysis, knowledge work — there is no meaningful functional difference between the two. The restrictions only apply to a narrow category of high-risk prompts.
Pricing: Is Claude Fable 5 Worth It?
Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens through the API. That is exactly double Claude Opus 4.8’s pricing ($5/$25 per million tokens). It is also double GPT-5.5 on input.
For subscription users on Claude Pro, Max, Team, or seat-based Enterprise plans, Fable 5 counts as 2x usage. It was available at no extra cost through June 22, 2026. After June 23, it requires usage credits until Anthropic can expand capacity to make it a standard plan feature.
Prompt caching is available with a 90% input token discount, which significantly reduces costs for repetitive or long-context workloads.
How to Access Claude Fable 5
There are several ways to get started with Claude Fable 5:
- Claude API: Use model ID
claude-fable-5directly in your API calls via the Claude Console - Claude.ai: Available on Pro, Max, Team, and Enterprise plans (requires usage credits after June 22)
- Claude Code: Integrated with Anthropic’s command-line agentic coding tool
- Amazon Bedrock: Available through AWS
- Vertex AI: Available on Google Cloud
- Microsoft Foundry: Available for enterprise Microsoft customers
- GitHub Copilot: Available to Copilot users (note: admin policy off by default)
Safety and Controversy
The release of Fable 5 was not without turbulence. Shortly after launch, Anthropic faced criticism for covertly limiting the model’s ability to be used for AI research — a policy that was widely described as “secret sabotage” of the model’s capabilities. Anthropic walked back that policy within 24 hours following backlash from researchers and developers.
On the security side, Anthropic says it stress-tested Fable 5’s safety classifiers through an external bug bounty program that logged over 1,000 hours of jailbreak attempts without finding any universal jailbreak. External red-teaming organizations also failed to break the safety classifiers universally before the model’s public launch.
The broader context matters too. Anthropic released this model while simultaneously warning major global AI labs about the risks of rapid AI advancement, including the possibility of systems achieving recursive self-improvement (RSI) — the ability to autonomously improve themselves without human guidance.
Who Should Use Claude Fable 5?
Claude Fable 5 is best suited for:
- Software engineers working on large, complex codebases who need multi-step autonomous task completion
- AI developers and researchers building products that require the strongest available base model
- Enterprise teams running long-horizon knowledge work tasks that previous models couldn’t sustain over time
- Data scientists who need extended thinking and high-output responses for complex analysis
For everyday writing, summarization, or simple Q&A, Claude Opus 4.8 or Claude Sonnet 4.6 remain the more cost-effective choices. Fable 5’s advantage becomes clear on tasks that are long, ambiguous, multi-step, or require deep reasoning across massive amounts of context.
Final Verdict
Claude Fable 5 is a genuine capability jump, not an incremental model refresh. The SWE-Bench Pro lead is substantial, the 1 million token context window opens up use cases that were impractical before, and real-world demonstrations like the Stripe codebase migration show that the benchmark numbers translate to actual productivity gains.
The price is steep at $10/$50 per million tokens, and the mandatory 30-day data retention may be a blocker for privacy-sensitive enterprise workflows. But for teams that need the best available model for serious engineering and knowledge work, Fable 5 currently sits at the top of the leaderboard.
If you want to try it today, grab an API key from the Claude Console and call it with model ID claude-fable-5. Subscription users can access it through Claude.ai on Pro and higher plans.