What Claude Opus 5 means for you

Anthropic has released a new AI model called Claude Opus 5, and the headline change is value: it performs close to the company's most capable model at roughly half the price.

Anthropic launched Claude Opus 5 on 24 July 2026. It is now the default model for people paying for Claude Max, Anthropic's top subscription tier, and the strongest model available to Claude Pro subscribers.

You can reach it through the Claude app, Claude Code (Anthropic's coding tool), the Claude API, or through Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry.

Pricing stays the same as the previous Opus model: $5 per million "input tokens" (roughly, the words you send in) and $25 per million "output tokens" (the words it sends back). That works out at about half the cost of Claude Fable 5, Anthropic's other recent flagship. It can also handle a context window, the amount of text it can consider at once, of up to 1 million tokens, and produce responses up to 128,000 tokens long. A new "effort" setting lets you choose low, medium or high, trading cost against how hard the model works on a given task.

On benchmark tests, Opus 5 beat Fable 5 on 8 of 13 measures. On one test covering physics, chemistry and cryptography problems, it scored 9.7 percentage points higher than Fable 5, reaching 43.3%.

Anthropic also says this is its safest model yet. In internal testing, it scored zero on a benchmark measuring whether the model could be tricked into exploiting security vulnerabilities. It can search code for security flaws but is built to refuse scanning methods commonly used by attackers, and its safety "guardrails" now trigger 85% less often than they did with Fable 5.

Two new features arrive for developers: you can now interrupt the model mid-task and swap its tools, and prompts that get blocked by safety filters can instead be automatically rerouted to a different model rather than simply failing.

Opus 5 is Anthropic's fourth model launch in under two months, following Mythos 5, Fable 5 and Sonnet 5 in June. Before the official release, a test version codenamed "Claude Honeycomb EAP" briefly appeared in the coding tool Cursor's model picker on 9 July.

If you already pay for Claude Max or Pro, you get access to a more capable model at no extra cost, since it's now the default. For anyone building with the API, the lower relative price compared with Fable 5 means you can get similar or better results for less money.

The safety improvements matter most if you use Claude for coding: the model is designed to help you find security flaws in your own code while refusing to help exploit them in someone else's.

Claude Max and Claude Pro subscribers can access Opus 5 directly in the Claude app now. Developers can call it via the Claude API using the model ID claude-opus-5, or through Amazon Bedrock, Google Cloud Vertex AI or Microsoft Foundry. More detail is in Anthropic's announcement, reported by SiliconANGLE.