Astonishing Achievements by Claude Fable 5: A Documented List of Its Top Capabilities

When Anthropic launches a model and openly states it is "too powerful to ship without safety classifiers actively redirecting some of its responses," you are facing an unusual event. That is exactly what happened with Claude Fable 5, the first Mythos-class model made available to the public on June 9, 2026, which the company says exceeds any model it has ever made generally available, achieving state-of-the-art results on nearly all capability benchmarks. Below is a list of its most astonishing documented achievements — all announced figures and facts, not exaggerations.

1. A Codebase Migration From Two Months to a Single Day

The standout achievement came from the payments company Stripe. In early testing, Fable 5 performed a codebase-wide migration in a 50-million-line Ruby codebase in a single day. That work would have taken a full team more than two months by hand. Compressing months of engineering work into days is the clearest example of the model's ability to work autonomously for longer than any previous Claude model.

2. Topping Coding Benchmarks by a Wide Margin

The coding benchmark numbers are genuinely striking. Fable 5 scored 80.3% on the difficult SWE-Bench Pro, leading its nearest competitors by about 11 points and surpassing GPT-5.5, which scored 58.6%. On Cognition's FrontierCode Diamond benchmark, which measures high-quality agentic coding, it scored 29.3% against 13.4% for Claude Opus 4.8 and just 5.7% for GPT-5.5. More cleverly, it tops this benchmark even at "medium reasoning effort," meaning it may deliver stronger coding results without always needing maximum compute.

3. Beating Pokémon With Vision Alone

Among the most amusing achievements, and the most telling of the leap in computer vision: Fable 5 completed Pokémon FireRed relying on raw game screenshots only, with no maps, no game-state information, and no helper scaffolding. Previous models needed additional support tools just to make any progress on the same task. The achievement is not about gaming itself, but about what it reveals: an ability for visual perception and sequential decision-making from raw visual input.

4. Rebuilding an App's Source Code From an Image

On the vision front too, the model became the new state-of-the-art for visual tasks. It can rebuild a web app's source code from screenshots alone, and extract precise values from scientific figures. This capability bridges a gap that long separated "seeing" an interface from "reproducing" it in code.

5. Breaking the 90% Barrier in Complex Analysis

In knowledge work, the company Hex reported a ten-point jump compared to Opus 4.8 on complex analytical tasks, with Fable 5 breaking the 90% barrier on its core benchmark for the first time. Cursor also described it as opening "a class of long-horizon problems that were out of reach for earlier models."

6. Novel Scientific Hypotheses and Protein Design

Here precisely is why the caution appears. The more powerful Mythos 5 version (the same model with safeguards lifted, available only to limited partners) recorded breakthroughs in drug design and molecular biology, described by Anthropic as the first of its models to consistently produce novel, compelling scientific hypotheses. In health, the model scored 66.0% on HealthBench Professional against 56.9% for Opus 4.8.

The Other Side: Why Does "Terrifying Efficiency" Come With Safeguards?

Honesty requires noting the other side, which is an integral part of the story rather than a footnote. Because of these capabilities, especially in cybersecurity, Anthropic launched the model with a layer of safeguards that redirect sensitive questions (in cybersecurity, biology, and chemistry) to the less-capable Opus 4.8, with the user informed. These safeguards are tuned conservatively, so they sometimes catch harmless requests, but they trigger in less than 5% of sessions; that is, more than 95% of sessions actually run at the model's full capability. In practice, this means roughly one session in twenty may not be running on the model you chose — a consideration worth noting before building a production workflow on it.

Practical Considerations

Fable 5 comes at roughly double the price of Opus 4.8 (10 dollars per million input tokens and 50 for output), and is available within the Pro, Max, Team, and Enterprise subscription plans through June 22, after which it requires usage credits. Its real strength shows in long, complex tasks, where its lead widens the longer and more complex the task gets, while it may not be the most economical choice for simple tasks.

Conclusion

Claude Fable 5's list of achievements is genuinely impressive without any hype: a two-months migration in a day, topping coding benchmarks by a wide margin, beating a game with vision alone, and novel scientific hypotheses. But the most important thing in the story is not the numbers alone, but that the model's capabilities are large enough that its maker releases it with safeguards that voluntarily curb some of its responses. That is the real "terrifying efficiency": not in exaggerated fantasy, but in a documented leap that makes talk of safety an inseparable part of talk of capability.