Anthropic claims latest model beats Gemini and GPT for coding

Anthropic CEO Dario Amodei / Photo credit: Mijansk786 / Shutterstock

Anthropic has unveiled Claude Opus 4.5, bringing its top-of-the-line AI model up to speed with its other recently launched models, Sonnet 4.5 and Haiku 4.5.

The US-based AI lab says that Claude Opus 4.5 is its “most intelligent” and “safest model” to date, positioning it as the model of choice for coding and agentic workflows. It also serves as a direct upgrade to Sonnet 4.5.

The release follows recent launches from Anthropic’s rivals. OpenAI rolled out a minor update in the form of GPT-5.1 on November 13, while Google released Gemini 3 on November 19 after making significant improvements to its predecessor.

In a blog post, Anthropic highlights Claude Opus 4.5’s edge over GPT-5.1 and Gemini 3 Pro on several benchmarks, primarily those that measure agentic coding (SWE-bench) and agentic tool use performance (Tau2-bench).

The company says that for coding use cases, Claude Opus 4.5 can handle “ambiguity and reasons about trade-offs without handholding.” In practice, this means the model can solve complex bugs and complete tasks that are too difficult for Sonnet 4.5.

When used to automate office tasks, Claude Opus 4.5 can independently refine the agents’ capabilities by storing insights from past work and applying them to new challenges.

As a result, Claude Opus 4.5 agents can reach peak performance after four iterations, while others can’t match its quality even after 10 iterations, notes Yusuke Kaji of Rakuten, who had early access to the model. Kaji is the general manager of AI for business at the Japanese tech conglomerate.

Another upgrade is in terms of API costs. The previous version of Opus was notably expensive to use at US$15 per million input tokens and US$75 per million output tokens. In comparison, Claude Opus 4.5 is priced 3x cheaper.

Alongside the new model, Anthropic has expanded the availability of its AI assistants. Claude for Chrome, which allows the chatbot to live inside Google’s browser and navigate autonomously on behalf of users, is now available to subscribers of Claude Max, the company’s premium tier.

On the other hand, Claude for Excel now provides support for pivot tables, charts, and file uploads on Microsoft’s spreadsheet program, and is available to Max, Team, and Enterprise users. Both tools are powered by Claude Opus 4.5.

Anthropic is also updating Claude’s desktop app to include a dedicated mode for Claude Code, its agentic coding tool. This allows Claude Code users to run multiple local and remote sessions in parallel.

In addition, the desktop app has fixed the context window limit problems that users frequently complain about. It can now automatically summarize earlier context as needed, which means long conversations will no longer hit a wall.

See also: Will India’s trade tensions spill over to OpenAI, Anthropic?