Anthropic Launches Claude Opus 4.5 With Major Upgrades in Coding and Efficiency

Anthropic released Claude Opus 4.5, the company's frontier artificial intelligence (AI) model, on Monday. The final member of the Claude 4.5 family and the most performant model in the series comes with major improvements in code generation, long-context reasoning, and agentic capabilities. At the same time, the AI firm highlighted that the model uses fewer tokens to complete complex tasks. Opus 4.5 is available via the Claude API and on supported cloud platforms, replacing Opus 4.1 as Anthropic's top-tier commercial model.

What's New With Claude Opus 4.5

In a newsroom post, Anthropic announced and detailed the new large language model. The company says the Claude Opus 4.5 brings marked improvements in three areas of coding, agentic tool use, and long-context reasoning. On coding, the company reports that Opus 4.5 solves more long-horizon coding tasks than its predecessor while using up to 65 percent fewer tokens.

In practice, this means the model can process the same instructions using less computational cost and within smaller prompt windows. The company attributes this improvement to better planning and more efficient internal reasoning steps.

Opus 4.5 is trained to manage more complex multi-step workflows. During internal tests, Anthropic claimed the model was able to refactor two separate codebases at once, coordinate the work of three agents, and maintain high-level plans while executing low-level details. These capabilities are enabled by improving long-context reasoning of the model and a new system to improve tool calling (instead of pre-loading a large library, the LLM only calls tools relevant to the task). This is said to reduce context usage by up to 85 percent.

For content generation, Opus 4.5 is designed to handle long documents more reliably. The model is capable of producing multi-page narrative chapters (10–15 pages in Anthropic's example) and maintaining characters, plot direction and tone across those longer passages. The company also claims better performance in complex 3D reasoning exercises, where the model describes scenes or spatial relationships in greater detail than earlier versions.

Coming to benchmarks, the company conducted internal testing and claimed that Claude Opus 4.5 outscored rivals in code-based tests. Notably, in the SWE-Bench Verified benchmark, which measures agentic coding, Opus 4.5 was said to score 80.9 percent, compared with Gemini 3 Pro at 76.2 percent and GPT-5.1 Codex Max at 77.9 percent.

Anthropic has also made Claude Opus 4.5 more affordable than its predecessors. The company says the model offers similar or improved performance with roughly one-third of the cost for many enterprise workloads. The LLM is currently available across Claude's app and website to paid subscribers, via Anthropic's application programming interface (API), as well as major cloud platforms, including Google Vertex AI and Amazon Bedrock.