Skip to content
AnthropicGA

Claude Opus 4.6

The reasoning + long-context flagship. THE model for high-stakes synthesis.

Context

1M

Max output

128K

Input /1M

$5.00

Output /1M

$25.00

Live pricing via OpenRouter

Best for

  • Abstract reasoning (#1 ARC-AGI-2, 68.8%)
  • Computer-use agents (#1 OSWorld, 72.7%)
  • 1M-context research synthesis and long agent sessions
  • Parallel agent orchestration (Agent Teams)

Watch out

Premium pricing. For high-volume routine steps, route to a cheaper model.

For creators. The reasoning brain behind serious creator OS builds and multi-file code generation.

Benchmarks

terminal bench 265.4
arc agi 268.8
osworld72.7
biglaw bench90.2
mrcr v2 1m76

Capabilities

  • Adaptive thinking (auto-determines reasoning depth)
  • 1M token context window (beta)
  • 128K output tokens (2x previous)
  • Agent Teams (parallel Claude Code agents)
  • Compaction API (server-side context summarization)
  • Enhanced agentic coding and debugging
  • Lowest over-refusal rate in Claude family
  • Fine-grained tool streaming (GA)
  • Data residency controls (US-only option)

Compare Claude Opus 4.6

Gemini 3.5 Flash vs Claude Opus 4.6

Different tiers, different jobs. Flash wins cost-sensitive agentic coding (76.2% Terminal-Bench 2.1 at a fraction of the cost); Opus 4.6 wins high-stakes reasoning (68.8% ARC-AGI-2) and computer-use.

Claude Opus 4.6 vs GPT-5.2 Pro

Opus 4.6 for reasoning and long-context depth; GPT-5.2 Pro for native voice and the broadest multimodal + integration footprint.

More from Anthropic

Claude Opus 4.5Claude Sonnet 4.5Claude Haiku 4.5

Sources