Anthropic's Claude Opus 5.5 Costs 40 Percent Less to Run

Anthropic's new flagship AI model is Claude Opus 5.5, released on September 22, and the headline number is the price. It performs at roughly the level of the company's top-tier Claude Fable 5.1 on most work, but Anthropic said it costs about forty percent less to run on typical workloads than its predecessor, Opus 5, according to Reuters.

The list price dropped as well. Reuters reported that the model costs four dollars per million input tokens and twenty dollars per million output tokens, twenty percent below Opus 5. For developers running long agent sessions, Anthropic also cut cache reads, which dominate the cost of agentic coding, to twenty cents per million tokens, according to AI Weekly's summary of the announcement.

Faster output and stronger benchmark scores

Output arrives more than thirty percent faster than Opus 5, Anthropic said, and Claude Opus 5.5 leads the company's internal benchmarks in agentic coding, computer use, and knowledge work. On Terminal-Bench 4.0, which tests real software tasks, Anthropic reported a score of sixty-six point four percent for Claude Opus 5.5, ahead of fifty-five point eight percent for Fable 5.1. CNA reported that the new model also outscored OpenAI's GPT-5.6 Sol on a software development benchmark while costing roughly a third as much to run.

The company also said Claude Opus 5.5 communicates more naturally than Opus 5, putting key information up front and using less jargon, a change MacRumors noted addresses a common complaint about the earlier model. In early testing cited by Anthropic, one user finished a migration of hundreds of thousands of lines of code in under a day, work the company said would have taken an engineering team weeks. Anthropic cautioned, though, that benchmark margins are a less reliable guide at this level of capability, and said the real-world gap between Claude Opus 5.5 and Fable 5.1 is narrower than the scores suggest. Knowledge work is where these gains land hardest for ordinary users, and our earlier reporting found that AI workplace productivity is up while strategy lags behind.

Safety testing gets tougher

This is Anthropic's first major release since chief executive Dario Amodei urged AI companies to slow down the pace of releasing new capabilities over safety concerns, according to CNA. Before launch, Claude Opus 5.5 went through external testing by the independent safety research groups Frontier Design and METR, and it ships with safeguards previously reserved for the company's most capable systems. That slowdown debate has real teeth: a lawsuit we covered earlier raised the question of whether the biggest AI labs coordinated on an AI slowdown.

On internal safety testing, Reuters reported the model was eighty-five percent less likely than Opus 5 or the Mythos system to attempt to bypass its prescribed boundaries, and Anthropic said it posted its best scores to date on an automated behavioral audit covering nearly two thousand scenarios. AI Weekly reported new safeguards including a "preserved thinking" protection against model distillation, a fallback to the less capable Opus 4.8 for certain cybersecurity tasks, and biology capabilities gated behind a Life Sciences Verification Program. AI systems probing real infrastructure is not a theoretical worry; we covered a recent test in which Gemini broke into three real companies.

What comes next

Claude Opus 5.5 is available now on Amazon Web Services, Google Cloud, and Microsoft Azure. MacRumors reported that subscribers on Pro, Max, Team, and seat-based Enterprise plans received higher five-hour usage limits along with a rate-limit reset they can use any time before October 22. Anthropic also said Claude Sonnet 5.5 and Claude Haiku 5.5 will arrive in the coming weeks, bringing many of the same performance, speed, and safety improvements to the rest of the model family.

The timing is pointed. Just before the launch, PYMNTS reported that a growing number of software companies are turning to open-weight AI models, including Chinese ones, because of the cost of proprietary models from US developers. A flagship-class model at a fraction of the price is Anthropic's answer to that pressure. The company also plans to give vetted cybersecurity and life-sciences researchers broader access to the model, according to CNA, which fits the pitch: cheaper, safer frontier AI, released into a market that is suddenly very price sensitive.