Anthropic announced Tuesday that Claude Fable 5, the first publicly accessible iteration of its Mythos AI system, is now available via the Claude API and the company’s consumption‑based Enterprise plans. The rollout marks a shift from the limited preview that began in April, when Mythos was offered only to a handful of partners due to cybersecurity concerns. This time, Anthropic is targeting hundreds of organizations across 15 countries, especially those that manage critical infrastructure.
Fable 5 is billed as a step up in software engineering, knowledge work, and vision capabilities. Yet Anthropic has paired the performance boost with hard safety limits. In domains it deems high‑risk—such as cybersecurity, biology, chemistry and distillation—the model blocks responses and automatically defers to its predecessor, Claude Opus 4.8. Internal testing showed that such deferrals are rare; early data indicate that roughly 95 % of sessions run entirely on Fable 5 without falling back.
Access to the model follows a two‑phase schedule. Through June 22, Fable 5 is included at no extra charge in Pro, Max, Team and seat‑based Enterprise plans. Starting June 23, the model will be removed from those plans, and users will need to purchase usage credits. Anthropic says it plans to restore Fable 5 as a standard subscription feature as soon as possible.
Safety measures extend beyond content blocking. The company now requires a 30‑day retention period for all traffic, even for enterprises that previously had zero‑retention agreements. Anthropic will not use retained data for training; instead, it will analyze it to defend against complex attacks, identify novel jailbreak attempts and reduce false positives. The firm claims its internal bug‑bounty program and external red‑team assessments failed to produce any universal jailbreaks after more than 1,000 hours of testing.
Third‑party benchmarks show the model delivering strong results. Analytics firm Hex gave Fable 5 a 90 % score on its core analytics benchmark for complex, long‑running tasks, noting “strong judgement and attention to nuance” on the hardest questions. Development platform Base44 highlighted the model’s ability to “one‑shot” full applications and praised its tool‑calling. Genspark, an AI‑powered workspace and agent platform, reported that Fable outperformed every other model in its evaluations, particularly on UI design and game coding tasks.
Pricing reflects the model’s advanced capabilities. Anthropic charges $10 per million input tokens and $50 per million output tokens, double the rates for Claude Opus 4.8. While the cost may deter some users, companies like shopping rewards platform Rakuten see value in the model’s self‑validation features. Rakuten’s statement said, “At the highest effort, Fable reflects on and validates its own work. For us, that’s what makes highly autonomous operations possible — the extra thinking pays for itself.”
Anthropic’s move comes as the firm prepares for a public market debut, joining rivals OpenAI and Elon Musk’s SpaceX in the race to commercialize frontier AI. The company also issued a broader appeal for the industry to adopt coordinated “brake‑pedal” mechanisms, warning that rapid advances could soon enable recursive self‑improvement without human oversight. By pairing a powerful new model with stringent guardrails and data‑retention policies, Anthropic hopes to set a precedent for responsible AI deployment as the technology reaches wider audiences.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.