Claude Opus 4.6 by Anthropic: first Opus with 1M token context, adaptive thinking, effort controls, context compaction, and lead performance on agentic search and knowledge work.
Capabilities, design details, and architectural traits
Claude Opus 4.6 is the Opus generation that introduced three capabilities new to the Opus class simultaneously: adaptive thinking, effort controls, and a 1M token context window in beta. These were not incremental updates to existing features - they each debuted with Opus 4.6. The model is designed around sustained, long-horizon agentic work, with documented improvements in planning, tool use, self-correction during code review, and reliability across large codebases.
| Trait | What it means for Opus 4.6 specifically |
|---|---|
| Adaptive thinking - first Opus deployment | thinking: {type: "adaptive"} was introduced with Opus 4.6; prior Opus models only had binary extended thinking on/off |
| Effort controls - four levels | low, medium, high (default), and max effort levels debuted here; Anthropic explicitly documents that high may overthink simple tasks and recommends dialing down via /effort |
| 1M token context - first Opus | Beta 1M token context window arrived with Opus 4.6; previous Opus models topped at 200k |
| Context compaction - first general availability | Automatic context summarization near window limits launched alongside Opus 4.6 on the API |
| ASL-3 deployment with published Sabotage Risk Report | Opus 4.6 is the first model for which Anthropic committed to and published a formal Sabotage Risk Report under its Responsible Scaling Policy, separate from the system card |
| Agent teams in Claude Code - debut | The ability to assemble multi-agent teams within Claude Code launched with Opus 4.6 |
The Opus 4.6 system card documents that the model roughly reached pre-defined ASL-4 rule-out thresholds on benchmark tasks. Because a clean rule-out was no longer straightforward, Anthropic supplemented the system card with a dedicated Sabotage Risk Report - a formal safety artifact now required for all future models exceeding Opus 4.5's capability level under the Responsible Scaling Policy.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | Anthropic |
| Release Date | February 5, 2026 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | - | May 2026 | - |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $5 Best Input Pricing | $5 Best Input Pricing | $10 |
| Output Pricing | $25 Best Output Pricing | $25 Best Output Pricing | $50 |
| Modalities | |||
| Inputs | textimagefile | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 44.9 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | - | 78.0 Best Coding Index | 76.5 |
| Agentic Index | - | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
