American company Anthropic has expanded its artificial intelligence model lineup with Claude Opus 5. The new model is designed to combine high performance in programming, document processing, and autonomous agent management with significantly lower costs than the current flagship model Claude Fable 5.
Anthropic describes Claude Opus 5 as a significant generational leap in the Opus line, intended primarily for long-term and complex tasks, deeper reasoning, and professional work. Although Fable 5 remains the company's most capable widely available model, Opus 5 approaches or surpasses it in a number of tests at half the price.
The new model can simultaneously process both text and image inputs. It offers a context window of one million tokens and can generate a response up to 128,000 tokens long. This makes it possible to analyze extensive documentation, entire software projects, text databases, or large numbers of images within a single prompt.
Developers can choose from five reasoning intensity levels — low, medium, high, xhigh, and max. The default setting uses adaptive reasoning, in which the model adjusts the amount of compute to the difficulty of the task. Also included are tools for file handling, programming, web search, memory, computer control, and integration via the MCP protocol.
Top of the artificial intelligence benchmarks
Claude Opus 5 reached first place in Artificial Analysis's Intelligence Index upon its release. At the maximum reasoning level it scored 61 points, narrowly surpassing Claude Fable 5. It performed particularly strongly in tests of agentic work, multimodal reasoning, and solving complex knowledge tasks.
Its result on the ARC-AGI-3 benchmark also drew attention; this benchmark measures how quickly a model can learn the rules of a previously unknown interactive environment. Claude Opus 5 achieved a score of 30.2 percent according to published results, nearly four times higher than the previous best model. ARC-AGI-3 differs from standard knowledge tests in that it does not provide models with clear instructions and primarily evaluates their ability to discover rules, plan, and respond effectively to novel situations.
The model also recorded strong results in programming, workflow automation, and long-horizon task completion. According to Anthropic, it is better at distributing work among subagents, monitoring its own progress, and detecting errors during the planning phase rather than only after generating a final response.
Lower price, but still a premium model
Claude Opus 5 is priced at five dollars per million input tokens and 25 dollars per million output tokens via the API. Caching already-processed portions of a prompt costs 50 cents per million tokens. The price is the same as the previous Claude Opus 4.8, but half that of Fable 5.
Anthropic also offers a fast mode intended to deliver significantly quicker responses. This is available only through the Claude API and costs ten dollars per million input tokens and 50 dollars per million output tokens. The faster processing therefore comes at double the price.
Claude Opus 5 is available through the Claude API as well as via Amazon Bedrock, Google Cloud, and Microsoft Foundry cloud services. For Claude Max subscribers it has become the default model, while users of the cheaper Claude Pro plan receive limited access.
The new model addresses some of the objections users had to the more expensive Fable 5. Opus 5 is said to refuse ordinary scientific or safety-related queries less often and to offer a more balanced price-to-performance ratio. Fable 5, however, retains an advantage for certain tasks requiring precise factual recall or the highest available level of capability.
Anthropic thus continues its strategy focused on powerful and relatively expensive models for professional deployment. Claude Opus 5 is aimed primarily at developers, businesses, and users who need to process large documents, build software, or run autonomous agents over extended periods.
deeplearning.ai/gnews.cz - GH