Legacy (still served)
Released 2026-02-05; ID
claude-opus-4-6; 1M context / 128K output; retirement not sooner than 2027-02-05. Superseded by claude-opus-5 (via claude-opus-4-dot-7 and claude-opus-4-dot-8). Extended (manual budget) thinking is deprecated on this model; adaptive thinking is the mode.
Claude Opus 4.6
by Anthropic
Frontier-level reasoning model with expert coding, autonomous agents, and adaptive thinking for enterprise-scale AI operations
See https://www.anthropic.com/news/claude-opus-46
Features
- Frontier-level reasoning that sustains depth across complex problems
- Expert-level software engineering and code quality
- Autonomous agentic workflows with multi-tool orchestration
- 1M token context window (per Anthropic docs, 2026-09-30; earlier 200K/1M-beta wording was from launch)
- 128K max output tokens
- Adaptive thinking: dynamic allocation of reasoning based on task complexity
- Four effort levels (low/medium/high/max) for balancing intelligence, speed, and cost
- Context compaction (beta) for extended conversations and long-running agents
- Advanced computer use and visual understanding
- Edge case handling and self-correction
Superpowers
Claude Opus 4.6 is designed for professional teams operating AI systems at enterprise scale. Its frontier-level reasoning sustains across complex problems without requiring constant human intervention. The adaptive thinking feature automatically decides when deep reasoning is needed—optimizing for speed on simple tasks while enabling comprehensive analysis on hard problems.
Best for:
- Large codebase refactoring and modernization
- Autonomous agentic systems (research, engineering, product development)
- Code review and security analysis
- Enterprise knowledge work and multi-step document generation
- Extended conversations without context limits (context compaction beta)
- Complex reasoning tasks requiring edge case handling
- Multi-tool orchestration and parallel task execution
Effort Controls: Fine-grained performance tuning lets you balance intelligence and cost per task—use “low” for simple queries, “max” for research-grade reasoning.
Availability
- Azure Foundry: Integrates with M365 Work IQ, Fabric IQ, web access
- Google Cloud Vertex AI
- Amazon Bedrock
- Kiro: 2.2x credit multiplier
- Anthropic API: Direct access
Pricing
Enterprise-grade variable pricing based on token usage and effort level. Lower effort levels reduce costs. Contact Anthropic for volume licensing and custom arrangements.
Sources
- https://platform.claude.com/docs/en/about-claude/model-deprecations (accessed 2026-09-30)
- https://platform.claude.com/docs/en/models/overview (accessed 2026-09-30)