Claude Opus 4.6
Anthropic's largest Claude model, built for long agentic coding and knowledge work
Claude Opus 4.6 is Anthropic's flagship Claude model, released in 2026 as an upgrade to Claude Opus 4.5. It targets coding, agentic tasks and professional knowledge work, and is the first Opus-class model to offer a 1 million token context window in beta.
- First Opus-class Claude with a 1M token context window, in beta
- Highest score reported on Terminal-Bench 2.0 agentic coding
- Beats GPT-5.2 by about 144 Elo on GDPval-AA, per Anthropic
- API pricing unchanged at $5/$25 per million tokens
- Adaptive thinking and an /effort parameter control reasoning depth
What it is
Claude Opus 4.6 is a general-purpose language model from Anthropic, positioned as the company's smartest Claude. It is aimed at coding, agentic work and everyday professional tasks such as financial analysis, research, and working with documents, spreadsheets and presentations. The company says it improves on Claude Opus 4.5 in planning, code review, debugging and reliability inside large codebases. It is the first Opus-class model with a 1 million token context window, offered in beta. Anthropic says the model performs strongly on agentic coding, deep web search, multidisciplinary reasoning and economically valuable knowledge work.
How it came about
Anthropic builds Claude models with Claude itself: engineers use Claude Code daily, and each new model is first tested on the company's own work. Opus 4.6 arrived as an upgrade to Claude Opus 4.5, with Early Access partners testing it before general release. The company says the model focuses on the hardest parts of a task without prompting, moves faster on routine parts, and stays productive over longer sessions. It launched alongside product updates including agent teams in Claude Code, context compaction on the API, adaptive thinking, effort controls, upgrades to Claude in Excel, and Claude in PowerPoint in research preview.
Inside the model
Anthropic has not published parameter counts or training data details for Opus 4.6. The company describes it as planning more carefully, revisiting its reasoning before answering, and sustaining agentic tasks over longer runs. On the API it supports compaction, where the model summarises its own context to keep working past normal limits. Adaptive thinking lets it decide from context how much extended thinking to use, and an /effort parameter lets developers trade intelligence against speed and cost; the default is high, and Anthropic recommends medium when the model overthinks simple tasks. In Claude Code it can assemble teams of agents working in parallel.
What it can do
Anthropic reports the highest score on Terminal-Bench 2.0, an agentic coding evaluation, and says it leads other frontier models on Humanity's Last Exam, a multidisciplinary reasoning test. On GDPval-AA, which measures economically valuable knowledge work in finance, legal and other fields, the company says it beats OpenAI's GPT-5.2 by about 144 Elo points and Claude Opus 4.5 by 190 points. It also leads on BrowseComp, which tests finding hard-to-locate information online. Partner reports cite strong results in legal reasoning, cybersecurity investigations, code review and long-context research work. Deeper thinking can add cost and latency on simpler problems.
How to use it
Claude Opus 4.6 is available through claude.ai, the Claude API under the model name claude-opus-4-6, and major cloud platforms. API pricing is $5 per million input tokens and $25 per million output tokens, unchanged from the previous model. The weights are not released; access is through Anthropic's products and partners.
Safety and testing
Anthropic says its system card shows an overall safety profile as good as or better than other frontier models, with low rates of misaligned behaviour across safety evaluations. The company published extensive safety testing alongside the release. No specific incidents are reported here.
Reception and debate
Early Access partners described the model as Anthropic's strongest, praising its planning, autonomy and handling of large codebases, with some reporting gains in bug catching, legal reasoning and long-context consistency. One partner said it completed a multi-million-line codebase migration in half the expected time. These are company-published testimonials rather than independent reviews.
Why it matters
Anthropic presents Opus 4.6 as extending the frontier for agentic coding, deep search and expert-level reasoning. It is used inside Anthropic's own engineering work and shipped into products including Claude Code, Cowork, Claude in Excel and Claude in PowerPoint, as well as partner tools for legal, security, design and software development work.
Questions people ask about Claude Opus 4.6
What is Claude Opus 4.6?
It is Anthropic's most capable Claude model, released in 2026 as a successor to Claude Opus 4.5. It is aimed at coding, agentic tasks and professional knowledge work, and is the first Opus-class model with a 1 million token context window in beta.
How do I use it?
You can use it on claude.ai, through the Claude API as claude-opus-4-6, and on major cloud platforms. Anthropic prices it at $5 per million input tokens and $25 per million output tokens, the same as the previous model.
How large is it?
Anthropic has not published a parameter count for Opus 4.6. The company describes improvements in planning, code review, debugging and long-running agentic work rather than giving model size details.
How does it compare to earlier Claude models?
Anthropic says it improves on Claude Opus 4.5 in coding, planning and reliability in large codebases. On GDPval-AA, the company reports it beats its predecessor by 190 Elo points and OpenAI's GPT-5.2 by about 144 points.
What safety testing was done?
Anthropic published an extensive system card and says the model shows an overall safety profile as good as or better than other frontier models, with low rates of misaligned behaviour across its safety evaluations.
Good reading and sources
There is no English Wikipedia article on Claude Opus 4.6 yet, so the facts on this page come from the sources listed above: anthropic.com/news/claude-opus-4-6. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.