Rahul Somvanshi
If your business runs on Claude, or you’re weighing it against OpenAI, the main question today is simple: what changed, and does the price cut matter? Anthropic released Claude Opus 5.5 on Tuesday. It says the model performs close to its most powerful model, Claude Fable 5.1, on most work. It also says Opus 5.5 costs 40% less to run than Opus 5 on typical workloads. It passed outside safety checks before release.
Anthropic launched Claude Opus 5.5 on September 22, 2026, as the first model in its new 5.5 family. The list price is $4 per million input tokens and $20 per million output tokens. That is 20% below Opus 5’s $5 and $25.
Cached input costs $0.20 per million tokens, 60% less than before. Cached input is text the system has already processed and can reuse. Anthropic also says Opus 5.5 produces output about 30% faster than Opus 5. The model has a 1 million token context window by default and a maximum output of 128,000 tokens.
The model is available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Foundry. Developers call it with the identifier `claude-opus-5-5`.
The 20% list-price cut is the smaller half of the saving. The bigger claim is the 40% lower running cost on typical workloads. Anthropic puts that down to the model using fewer tokens and cheaper cached reads. A team spending $10,000 a month on Opus 5 could spend about $6,000 for similar work, if its usage matches Anthropic’s “typical” profile. Teams with unusual workloads should test before assuming the full saving.
Claude Opus 5.5 safety testing
Anthropic says two independent research groups, Frontier Design and METR, evaluated the model before launch. METR tests AI systems for risky autonomous behaviour.
Inside the company, Opus 5.5 scored better than any earlier Claude model on Anthropic’s automated behavioural audit. The audit runs nearly 2,000 simulated scenarios. Anthropic reports an 85% drop in attempts to get around boundaries compared with Opus 5.
Security firm Gray Swan tested the model against prompt injection. That’s when hidden instructions in a web page or document try to hijack an AI’s behaviour. Opus 5.5 tied Fable 5.1 for the lowest attack success rate.
The testing follows an Anthropic report on four incidents during outside cybersecurity evaluations. In those, misconfigured test environments accidentally connected Claude models to the real internet. In one case, a model uploaded a malicious software package that 15 third-party systems installed. In others, models accessed or changed real company records they mistook for test targets.
Anthropic called the incidents “serious.” It said the models showed “willingness to take harmful actions in narrow pursuit of a task.” The company says it has since added pre-release tests for these behaviours and new live monitoring. It also brought in METR to investigate independently.
On capability, Anthropic reports 66.4% on Terminal-Bench 4.0, a test of command-line coding tasks. It also reports 81.8% partial success on OSWorld 2.0, which measures how well an AI operates a computer. One early tester completed a 680,000-line code migration in under a day, Anthropic says.
How does it compare with OpenAI? About 90 minutes after this launch, OpenAI released GPT-6 Sol and Luna at half the price of their predecessors. Our GPT-6 Sol and Luna report covers that side. Each company picks its own benchmarks, so the headline scores can’t be compared directly.
Pricing is in US dollars across all regions, so UK and Australian businesses pay the same list rates. More coverage sits on our artificial intelligence and AI technology pages.
Claude Opus 5.5 launched on September 22 at $4 and $20 per million tokens. Anthropic says it costs 40% less to run than Opus 5, works 30% faster, has a 1 million token context window and passed outside safety testing by METR and Frontier Design. It is live on Anthropic’s platform, AWS, Google Cloud and Azure. Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks. Check back for updates.
—
