Anthropic today launched Claude Fable 5, the first publicly available version of its Mythos-class model, alongside Claude Mythos 5, the unrestricted enterprise version for critical infrastructure organizations. It is the most capable model the company has ever made available to the general public.
Fable 5 represents a significant leap over Claude Opus 4.8, which had been released just two weeks earlier. Where Opus 4.8 excelled in honesty and self-questioning, Fable 5 aims directly at autonomy and multi-step execution as its core differentiator.
What Fable 5 Brings
The model features a context window of 1 million tokens by default and can generate up to 128 thousand output tokens per request. On benchmarks, the numbers are striking:
- 80.3% on SWE-bench Pro, far surpassing GPT-5.5 (58.6%) and Gemini 3.1 Pro (54.2%)
- 29.3% on Frontier Code Diamond, nearly 6 times GPT-5.5 (5.7%) and more than double Opus 4.8 (13.4%)
- SOTA performance on vision, finance, and scientific reasoning benchmarks
Pricing stands at $10 per million input tokens and $50 per million output tokens, roughly double the cost of Opus 4.8. However, it is less than half the price of Claude Mythos Preview.
Key Capabilities
Autonomous software engineering. Stripe reported that Fable 5 compressed months of engineering into days. On a 50-million-line Ruby codebase, the model completed a full migration in a single day that would normally have taken an entire team.
State-of-the-art vision. Fable 5 can extract precise numbers from detailed scientific figures and reconstruct entire web applications from screenshots. In one demo, it played Pokémon FireRed from start to finish using only game screenshots, without maps or navigation aids.
Memory and autonomy. The model stays focused across millions of tokens in long-duration tasks. When given access to persistent file-based memory, its performance improved three times more than for Opus 4.8.
Guardrails and Safety
Fable 5 includes safety barriers that block responses in high-risk areas such as cybersecurity, biology, chemistry, and model distillation. In these categories, the model automatically redirects the query to Claude Opus 4.8.
These guardrails activate in fewer than 5% of sessions, according to Anthropic. The company describes the approach as “safety by design” — the most capable model needs the strictest protections.
Mythos 5, the enterprise version, has no such restrictions and is available to organizations managing critical infrastructure, in collaboration with the United States government.
Availability and Pricing
Fable 5 is available today on the Claude API and on consumption-based Enterprise plans. On subscriptions:
- Until June 22: included in Pro, Max, Team, and Enterprise plans at no additional cost
- Starting June 23: will require usage credits, with plans to restore it as a standard feature as soon as possible
What This Means
The launch of Fable 5 marks a turning point. Anthropic is implicitly saying that raw capability is no longer the problem — the problem is how to deploy it safely. The existence of Mythos 5 (unrestricted) for select partners and Fable 5 (with guardrails) for the general public creates a two-speed model that we will likely see replicated by other labs.
For developers, Fable 5 opens up real possibilities for long-running automation that previously required complex infrastructure. The price, while higher than Opus 4.8, remains competitive for tasks that once required entire human teams.