Anthropic rolled out Claude Fable 5 on Tuesday, marking the debut of a Mythos‑class language model that anyone with a paid subscription can access. The new offering mirrors the architecture of the company’s restricted Mythos system but adds a layer of safeguards designed to block queries in cybersecurity, biology, chemistry and other “distillation” scenarios. When a blocked request is detected, the system automatically switches to Claude Opus 4.8, delivering a safe answer instead of refusing outright.
According to Anthropic, the fallback mechanism activates in fewer than five percent of sessions, meaning most users will interact directly with the full‑capacity Fable 5 model. Early internal benchmarks show the model scoring more than ten percent higher than Opus 4.8 on select tests, a jump the company describes as “significant.” Independent evaluations have yet to be published.
Pricing reflects the performance boost. Anthropic charges $10 per million input tokens and $50 per million output tokens for Fable 5—double the rates for Opus 4.8, which launched at $5 and $25 respectively. To encourage early adoption, the company is offering the model at no extra cost to Pro, Max, Team and seat‑based Enterprise subscribers through June 22. After that date, users will need to purchase usage credits.
Early customers report higher return on investment, saying the model’s increased intelligence reduces the number of iterations required to complete a task. “You just get a higher ROI by having more intelligent models,” said Dianne Penn, Anthropic’s head of product management for research, in a CNBC interview. While promising, those efficiency claims are based on internal feedback and have not been independently verified.
Anthropic also refreshed its restricted offering, launching Claude Mythos 5 for the roughly 50 vetted partners in its Project Glasswing cybersecurity initiative. Unlike Fable 5, Mythos 5 lifts the cybersecurity and biology safeguards, keeping the model’s full power available only to trusted collaborators. Since April, Project Glasswing has identified more than 10,000 high‑ or critical‑severity vulnerability candidates across major software projects.
The timing of the launch is strategic. Anthropic filed a confidential IPO prospectus with the SEC last week, positioning the company for a possible public listing as early as this autumn. The filing follows a recent funding round that pushed the company’s post‑money valuation to $965 billion, briefly surpassing OpenAI’s $852 billion valuation after its own March round.
Industry analysts see the Fable 5 release as a concrete answer to a lingering question: how can Anthropic monetize Mythos‑class capabilities without exposing the risky toolsets that made the original model valuable—and dangerous? The answer, for now, is a model that is broadly more intelligent, carries a higher price tag, and includes safety filters that block a small fraction of potentially harmful queries.
Anthropic has also introduced a 30‑day data‑retention policy for all traffic generated by Fable 5 and Mythos 5. The data will be used for safety monitoring but will not feed back into model training, a move aimed at assuaging concerns about inadvertent data leakage.
With three of the most valuable private tech firms—Anthropic, OpenAI and Elon Musk’s SpaceX—poised to go public in the fall, the market could see more than $200 billion of new public‑market value. Anthropic’s latest model positions it to capture a slice of that upside by demonstrating that frontier AI can be delivered at scale while keeping misuse risks in check.
Dieser Artikel wurde mit Unterstützung von KI verfasst.
News Factory APP - agentische News für besseres SEO & AEO.