Anthropic Claude Opus 4.6 launched with 1 million token context window

On February 5, 2026, Anthropic launched Claude Opus 4.6, its latest and most advanced AI model. After months of anticipation and leaks buildup, the company upgraded its flagship model with improvements in intelligence and practical use.

Claude Opus 4.6 builds directly on its predecessor, Claude Opus 4.5. The new version plans more carefully, handles long-running tasks better, works reliably in large codebases, and catches its own mistakes through stronger code review and debugging. For the first time in an Opus-class model, it offers a 1 million token context window in beta, allowing it to process much larger amounts of information at once.

The model applies these abilities to everyday work. It runs financial analyses, conducts research, and creates or edits documents, spreadsheets, and presentations. In Anthropic’s Cowork feature, Opus 4.6 can manage multiple tasks on its own.

Performance benchmarks show clear gains. Opus 4.6 earned the highest score on Terminal-Bench 2.0, an agentic coding test, and leads all other frontier models on Humanity’s Last Exam, a difficult multidisciplinary reasoning challenge.

On GDPval-AA, which measures performance on valuable professional tasks in finance, law, and other fields, it outperforms OpenAI’s GPT-5.2 by about 144 Elo points and its own predecessor by 190 points. It also does better than any other model on BrowseComp, a test of finding hard-to-locate information online.

Anthropic Claude Opus 4.6

Long-context handling improved significantly. On a challenging needle-in-a-haystack test with 1 million tokens, Opus 4.6 scores 76%, far ahead of earlier models. This reduces “context rot,” where performance drops in very long conversations.

Additionally, Claude Code supports agent teams that work together on tasks. API features include context compaction for longer sessions, adaptive thinking that adjusts effort based on the problem, and simple effort controls to balance speed, cost, and intelligence. Claude in Excel received major upgrades, and Claude in PowerPoint launched in research preview.

Testing shows Opus 4.6 has low rates of unwanted behaviors like deception or excessive refusals, matching or improving on previous models.

Early users and partners report that Opus 4.6 feels more focused, thinks deeper on hard problems, and works more independently. Many describe it as a capable collaborator rather than just a tool.

Claude Opus 4.6 is available now on claude.ai, the Anthropic API, and major cloud platforms. Pricing stays the same at $5 per million input tokens and $25 per million output tokens.

(source)