Claude Opus 4.6: Anthropic's Most Capable Model Yet
Anthropic released Claude Opus 4.6 on February 5, 2026, and it's a significant leap forward. With a 1M token context window, 128K output tokens, and agent teams built into Claude Code, this model is redefining what's possible with AI-assisted development.
What's New
Opus 4.6 brings several major upgrades over its predecessor, Opus 4.5:
- 1M Token Context Window — A first for Opus-class models (in beta). This means you can feed entire codebases, lengthy documents, or massive datasets into a single prompt.
- 128K Output Tokens — Doubled from the previous 64K limit, enabling longer reasoning chains and more comprehensive responses.
- Agent Teams — Claude Code can now split large tasks across coordinated teams of agents, each handling a segment of the work in parallel.
- Adaptive Thinking — The recommended thinking mode, where Claude dynamically decides when and how much to reason through a problem. No more manually toggling extended thinking on and off.
- Context Compaction — For extended conversations, the model intelligently compresses prior context so you can keep working without hitting limits.
Benchmarks
The numbers speak for themselves:
| Benchmark | Result |
|---|---|
| GDPval-AA | Outperforms GPT-5.2 by ~144 Elo points |
| Terminal-Bench 2.0 | Highest score among frontier models for agentic coding |
| Humanity's Last Exam | Leads all frontier models |
| BrowseComp | Best in industry for information retrieval |
| MRCR v2 | 76% long-context retrieval (vs. 18.5% for Opus 4.5) |
That long-context retrieval jump from 18.5% to 76% is particularly impressive — it means the model can actually use information buried deep in massive contexts, not just accept them as input.
Coding Improvements
For developers, the improvements that matter most are:
- Better planning — Opus 4.6 thinks more carefully before acting, leading to fewer false starts on complex tasks.
- Larger codebase navigation — It can reliably work across large repos without losing track of file relationships.
- Stronger debugging — Code review and bug-finding capabilities have noticeably improved.
- Cybersecurity detection — Better at identifying vulnerabilities in code.
Pricing
Pricing stays the same as Opus 4.5:
- Standard: $5 input / $25 output per million tokens
- Premium (for prompts over 200K tokens): $10 input / $37.50 output per million tokens
Availability
Opus 4.6 is available on claude.ai, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry.
If you haven't tried it yet, the agent teams feature in Claude Code alone makes it worth exploring.