Overview
Claude Opus 4.6 is Anthropic's flagship Opus-tier model, released February 5, 2026. It sits at the top of Anthropic's three-tier lineup (Opus for the hardest work, Sonnet for everyday tasks, Haiku for speed and volume) and is aimed squarely at agentic coding, computer use, tool use, and complex multidisciplinary reasoning. At launch Anthropic positioned Opus 4.6 as an industry-leading model for these workloads, citing a top score on the Terminal-Bench 2.0 agentic-coding evaluation and a lead on Humanity's Last Exam.
Opus 4.6 ships with a 1M-token context window and up to 128K output tokens (extendable to 300K through the batch beta). It introduced adaptive thinking, letting the model decide when to engage deeper reasoning so it is fast and cheap on simple prompts but thorough on hard ones, alongside Anthropic's earlier extended-thinking mode. It accepts text, images, and PDFs and returns text, with strong long-context retrieval (76% on the MRCR v2 1M-token variant, up sharply from prior models).
Available on the Claude API (model ID claude-opus-4-6), claude.ai, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on Azure, Opus 4.6 is priced at $5 per million input tokens and $25 per million output tokens, with the full 1M-token window now billed at standard rates. It has since been succeeded by Claude Opus 4.7 and Opus 4.8, but remains available through the API as a legacy model.
| Released | 2026-02-05 |
|---|---|
| License | Proprietary (commercial, usage-based API) |
| Weights | API only |
| Parameters | Undisclosed |
| Context | 1M |
| Max output | 128K (up to 300K via batch beta) |
| Architecture | Undisclosed proprietary architecture. Supports adaptive thinking (the model decides when to reason more deeply) and extended thinking. Text and image input with text output. |
| Knowledge cutoff | Reliable knowledge cutoff May 2025; training data cutoff Aug 2025 |
| Modalities | Text, Vision, PDF |
| Status | Available (superseded by Opus 4.7/4.8; still served via API) |
Benchmarks
- SWE-bench Verified79.8%
- Humanity's Last Exam53%
- BigLaw Bench90.2%
- MRCR v2 (1M-token long-context retrieval)76%
- Terminal-Bench 2.0 (agentic coding)65.4%
- ARC-AGI-268.8%
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Pricing
| Input | $5 / 1M tokens per 1M tokens |
|---|---|
| Cached input | $0.50 / 1M tokens (cache hit/read) per 1M tokens |
| Output | $25 / 1M tokens per 1M tokens |
Full 1M-token context billed at standard rates (no long-context surcharge). 5-min cache write $6.25/1M, 1-hour cache write $10/1M. Batch API: $2.50 input / $12.50 output per 1M. Fast mode (research preview): $30 input / $150 output per 1M.
Strengths
- Frontier agentic coding and tool use, with a top score on Terminal-Bench 2.0 at launch
- 1M-token context window with strong long-context retrieval (76% on MRCR v2 1M variant)
- Adaptive thinking dynamically scales reasoning effort to balance speed, cost, and quality
- Strong expert-level reasoning, leading frontier models on Humanity's Last Exam at release
- Multimodal input (text, images, PDFs) for document analysis and visual question answering
- Full 1M context billed at standard per-token rates with no long-context surcharge
Best for
- Autonomous and agentic coding workflows (multi-step tasks, terminal/CLI agents)
- Long-context document analysis over large codebases, contracts, and research corpora
- Computer use and tool-calling agents
- Complex, multidisciplinary reasoning and research
- Enterprise legal, finance, and knowledge-work automation
- Visual question answering and analysis of charts, diagrams, screenshots, and PDFs
How to access
| Provider | Model ID |
|---|---|
| Anthropic Claude API ↗ | claude-opus-4-6 |
| Amazon Bedrock ↗ | anthropic.claude-opus-4-6-v1 |
| Google Cloud Vertex AI ↗ | claude-opus-4-6 |
| Microsoft Foundry (Azure) ↗ | claude-opus-4-6 |
Claude Opus — every version
The full lineage of the Claude Opus line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Claude Opus 4.8current | 2026-05-28 | 1M | Proprietary |
| Claude Opus 4.7 | 2026-04-16 | — | Proprietary |
| Claude Opus 4.6 | 2026-02-05 | — | Proprietary |
| Claude Opus 4.5 | 2025-11-24 | — | Proprietary |
| Claude Opus 4.1 | 2025-08-05 | — | Proprietary |
| Claude Opus 4 | 2025-05-22 | — | Proprietary |
| Claude 3 Opus | 2024-03-04 | — | Proprietary |
FAQ
When was Claude Opus 4.6 released?
Anthropic released Claude Opus 4.6 on February 5, 2026, available the same day on claude.ai, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on Azure. It has since been succeeded by Claude Opus 4.7 and Opus 4.8.
How much does Claude Opus 4.6 cost?
Standard pricing is $5 per million input tokens and $25 per million output tokens, with cache reads at $0.50 per million. The full 1M-token context is billed at standard rates. Batch processing is half price ($2.50 input / $12.50 output per million), and a fast-mode research preview is offered at $30 / $150 per million.
What is the context window and knowledge cutoff for Claude Opus 4.6?
Claude Opus 4.6 has a 1M-token context window and supports up to 128K output tokens (300K via the batch beta). Its reliable knowledge cutoff is May 2025, with a training data cutoff of August 2025.
Is Claude Opus 4.6 multimodal?
Yes. Opus 4.6 accepts text, images, and PDFs as input and returns text. Its vision capability handles charts, diagrams, screenshots, and photographs, making it useful for document analysis and visual question answering. It does not generate images, audio, or video.