Anthropic Calls Claude Opus 5.5 'Strongest-Performing Model We've Tested' as It Makes New Safety Claims

Anthropic has cut list prices by 20 per cent on its top Claude AI model while claiming new safety safeguards make it the 'strongest-performing model we've tested', as the company faces intensifying competition from OpenAI and rising Chinese rivals.

Claude Opus 5.5, launched on Tuesday, is the latest model in Anthropic's Opus line-up, above the mid-range Sonnet and cheaper Haiku tiers.

The San Francisco-based firm says the new model matches the performance of its top-tier Claude Fable 5.1 system on most tasks, at a fraction of the cost, while carrying safeguards previously reserved for its most advanced models in sensitive fields such as biology and cybersecurity.

Lower Prices Faster Output

Anthropic cut list prices for Opus 5.5 by 20 per cent compared with Opus 5, to $4 per million input tokens and $20 per million output tokens. Cache reads, which make up the bulk of costs in coding and agent-based work, fell by 60 per cent to 20 cents per million tokens.

On typical workloads, Anthropic said total costs fall by around 40 per cent because the model completes tasks using fewer tokens and produces output more than 30 per cent faster than its predecessor. Claude Pro, Max and Team subscribers will also get longer usage limits, along with a rate-limit reset they can save and use when needed.

Anthropic said Opus 5.5 outperformed OpenAI's GPT-5.6 Sol on a software-development benchmark while costing far less to run. In an internal trial, the company said, the model successfully found and fixed inefficiencies that cut page load times in 39 out of 40 runs.

New Safety Tests at Heart of Launch

The launch leans heavily on safety. Anthropic said its automated behavioural audit, which it calls its most comprehensive alignment test, gave Opus 5.5 the strongest results of any model the company has produced.

The firm said Opus 5.5 is markedly less likely than previous models to take hard-to-reverse actions or act outside agreed limits, and is more resistant to prompt injection than Opus 5.

According to Anthropic's own figures, the model was around 85 per cent less likely than Opus 5 or Claude Mythos 5.1 to attempt to bypass containment boundaries in a dedicated evaluation.

The safety framing follows chief executive Dario Amodei's recent call for AI developers to 'pace the frontier' by slowing capability releases so that safety work can keep up. Writing earlier this month, Amodei said addressing the risks of advanced AI 'requires even more prudence'.

Anthropic said Opus 5.5 was tested externally by independent groups Frontier Design and METR before release, in addition to its own alignment checks.

Because its biology and cybersecurity capabilities match those of Mythos 5.1 and Fable 5.1, the company said, Opus 5.5 is subject to the same restrictions, including limits on penetration testing and exploit generation, as well as safeguards around potentially dangerous biological research.

Anthropic is also widening access to Opus 5.5 for vetted cybersecurity and life-sciences researchers through its verification programmes.

Raised Stakes Ahead of Possible IPO

The release comes as Anthropic prepares for a potential initial public offering, with reports suggesting the company could seek a valuation approaching $2 trillion. The figure has not been confirmed by Anthropic.

The company was valued at $965 billion in its most recent private funding round. Anthropic's $65 billion Series H funding round in May gave it a post-money valuation of $965 billion.

Anthropic faces growing competition from OpenAI, which has also cut prices in recent months, and from open-weight models built in China and elsewhere, which are increasingly run on customers' own hardware rather than through paid subscriptions.

Anthropic said Sonnet 5.5 and Haiku 5.5 would follow in the coming weeks, with similar gains in performance, speed and safety.