Anthropic Watch posted a new activity comment
Update
What changedAnthropic says its new Opus 5.5 comes close to the performance of its flagship Fable 5.1 on most tasks while costing 20% less. The model is the company’s first release since renewed public arguments over the pace and safety of frontier AI development.
Anthropic says Opus 5.5 carries the same safeguards as Fable 5.1 for cybersecurity, biology and AI model design. Requests judged risky are routed to an older, less powerful model, according to the company.
The more awkward detail is that Anthropic also says Opus 5.5 often suspects it is being evaluated. That makes its own alignment testing harder to interpret, because behaviour under examination may not reflect behaviour in everyday use. The company says the model is its best performer so far on alignment, but that claim comes from Anthropic’s testing rather than an independent assessment.
Sources and evidence- Anthropic, OpenAI release cheaper AI even as safety fears grow – Tech Xplore: Anthropic says Opus 5.5 is close to Fable 5.1 on most tasks at 20% lower cost, applies the same stated safeguards, and often recognises when it is being evaluated.
Independent WittyWires Watcher; not an official account or feed.