Claude Opus 4.6: Features, Benchmarks, Pricing and What Changed

Anthropic released Claude Opus 4.6 on 5 February 2026, and it is comfortably the most capable AI model the company has ever shipped. This is not a minor version bump. Opus 4.6 brings a 1 million token context window, a completely reworked thinking system, doubled output limits, and benchmark results that put clear daylight between it and everything else on the market.

If you are building with the Claude API, using Claude Code professionally, or just trying to understand where the frontier sits right now, this is the model that defines it.

Key takeaway: Opus 4.6 scores highest on Terminal-Bench 2.0 (agentic coding), Humanity's Last Exam (reasoning), and GDPval-AA (professional knowledge work). It has a 14.5 hour task horizon, 1M token context, and 128K max output.

The 1M Token Context Window

Opus 4.6 moves beyond the standard 200,000 token limit to a massive 1M token context window. This is roughly 700,000 words—enough to hold an entire corporate document library or a complete codebase in memory.

Qualitative Shift in Accuracy

On MRCR v2 (extreme scale needle-in-a-haystack), Opus 4.6 scores 76%, compared to 18.5% for Sonnet 4.5. This effectively eliminates "context rot" for long-running sessions.

Adaptive Thinking Replaces Extended Thinking

Opus 4.6 introduces adaptive thinking, replacing the old manual budget system. The model now dynamically decides when and how much to think based on task complexity.

Old MethodManual budget_tokens. Wasted tokens or constrained reasoning.

New Adaptive ThinkingModel chooses reasoning depth. Controlled by effort levels (Low to Max).

Fast Mode: 2.5x Speed at a Premium

Available as a research preview, Fast Mode delivers up to 2.5x faster output generation. It's the same frontier intelligence running on optimized inference infrastructure.

Usage & Cost

Enable via speed: "fast" parameter. Ideal for real-time agent workflows and interactive coding.

Note: Fast Mode carries a premium price point ($30/$150 per MTok).

Benchmark Performance

Benchmark	What It Measures	Opus 4.6	GPT-5.2
Terminal-Bench 2.0	Agentic coding	65.4%	64.7%
GDPval-AA (Elo)	Knowledge work	1606	1462
ARC AGI 2	Problem solving	68.8%	54.2%

Breaking Changes and Migration

✕
Prefilling is gone: Assistant message prefilling is not supported on Opus 4.6. Migrating to structured outputs is required.
!
Output Tokens: Maximum output tokens doubled to 128K. SDKs require streaming for large requests to avoid timeouts.

Claude Opus 4.6 Pricing

Tier	Input (per MTok)	Output (per MTok)
Standard	$5 (£4)	$25 (£20)
Fast Mode	$30 (£24)	$150 (£120)