Claude Fable 5: Anthropic's Mythos-Class AI Is Now Available to Everyone

Anthropic made its boldest move yet on June 9, releasing Claude Fable 5 — the first publicly available model from its Mythos tier — to anyone with a Claude subscription. Previously, Mythos-class models were restricted to vetted government and enterprise partners. Now the most capable AI model Anthropic has ever built is just a login away.

The timing raises eyebrows. Just days earlier, Anthropic publicly warned that AI is "becoming too dangerous." Then it released this.

What Makes Fable 5 Different

Claude Fable 5 is not an incremental upgrade. It is a generational leap in AI coding capability.

On SWE-Bench Pro, the industry's gold-standard benchmark for real-world software engineering, Fable 5 scores 80.3%. That is 22 points ahead of OpenAI's GPT-5.5 (58.6%) and 26 points clear of Google's Gemini 3.1 Pro. On FrontierCode's Diamond split, it posts 29.3% — more than double Claude Opus 4.8's 13.4%. In document reasoning tasks, it hits 29.8% on the GDP.pdf benchmark versus GPT-5.5's 24.9%.

These are not abstract numbers. In a real-world demonstration, Fable 5 completed a 50-million-line Ruby codebase migration in a single day — work that had taken human engineering teams two months.

The model ships with a one-million-token context window and 128,000-token maximum output. It can sustain autonomous agentic work sessions stretching up to 12 hours, executing multi-page specifications without human intervention. Its knowledge cutoff is January 2026.

Developers Are Already Impressed

Early hands-on reports back up the benchmarks. Developer Simon Willison ran a roughly five-and-a-half-hour test session and came away calling the quality of Fable 5's API design, tests, code, and documentation "exceptional" — rating the output as several days of skilled human work. His bill: approximately $110.

In that session, Fable 5 autonomously upgraded a MicroPython WebAssembly project to full CPython via WASM and designed four LLM library features with built-in pause-resume mechanisms. Other demonstrations showed the model generating complete, playable video games from a single text prompt and clearing a full Pokémon FireRed playthrough using only raw screenshots — no navigation aids, no game state APIs, just vision.

The Safety Question

Anthropic's approach to releasing a Mythos-class model publicly is layered. Requests in high-risk domains — cybersecurity, chemistry, biology, and model distillation — automatically route to the more constrained Claude Opus 4.8. This fallback activates in fewer than 5% of sessions, according to Anthropic.

The company logged over 1,000 hours of red-teaming and reports no universal jailbreaks were discovered. The model executes its own responses more than 95% of the time.

Meanwhile, the fully unrestricted Mythos 5 variant remains locked down, available only to roughly 200 participants in Anthropic's Project Glasswing program — primarily cyberdefenders and infrastructure providers.

The tension is unmistakable: Anthropic tells the world AI is getting too dangerous, then hands the public the most powerful model it has ever built. The company's bet is that careful guardrails — automatic fallbacks, restricted domains, red-team-validated safety — can bridge the gap between capability and responsibility.

Pricing and Availability

Fable 5 is priced at $10 per million input tokens and $50 per million output tokens — double Claude Opus 4.8 but 50% cheaper than the original Mythos Preview pricing. Anthropic argues that token efficiency compensates: Fable 5 completes complex tasks in fewer steps, so the effective cost per task often comes out lower.

Claude Pro, Max, Team, and Enterprise subscribers get access at no extra cost until June 22. After that date, additional compute credits will be required. The model is also available through Amazon Bedrock for enterprise deployments.

Why It Matters

Claude Fable 5 marks the moment frontier AI stops being something that happens behind closed doors. An 80.3% score on SWE-Bench Pro means the model can handle the majority of real-world software engineering tasks that professionals face daily. A 50-million-line migration in a day means it is not just assisting developers — it is doing the work.

For the AI industry, the release intensifies the three-way competition between Anthropic, OpenAI, and Google. For developers, it is a tool available right now that demonstrably outperforms everything else on the market. And for everyone watching the safety debate, it is a test case: can you release the most powerful model in the world and keep it responsible?

Anthropic is betting yes. The next few weeks will show whether the public agrees.

---

Sources: TechCrunch, CNBC, Simon Willison, OpenTools, 9to5Google, About Amazon