← All models

Claude Opus 4.6

Published Publisher Anthropicclaude-opus-4-6

Overview

Claude Opus 4.6 is an Anthropic model for complex reasoning, agentic coding and long-running knowledge-work tasks. It was released on February 5, 2026. Anthropic now lists it as an active legacy model, with retirement no earlier than February 5, 2027. (anthropic.com)

Capabilities and uses

  • Accepts text and images and produces text. Its context window is 1 million tokens, with up to 128,000 output tokens per standard request. (platform.claude.com)
  • Supports coding, code review and debugging, multi-step agentic work, research, financial analysis, and work with documents, spreadsheets and presentations. (anthropic.com)
  • Supports adaptive thinking and effort controls. API context compaction can summarize older context for longer-running tasks. (anthropic.com)

Limitations

  • Outputs can contain errors and should be checked, particularly for consequential decisions or code changes. Longer reasoning can increase latency and cost; effort controls can help balance these trade-offs. (anthropic.com)
  • The documented reliable knowledge cutoff is May 2025; the training-data cutoff is August 2025. A large context window does not guarantee perfect retrieval or reasoning. (platform.claude.com)
  • At launch, the 1-million-token context was a beta feature with a long-context premium. Current Anthropic pricing documentation states that the full window is billed at standard per-token rates. Platform availability and charges may differ. (anthropic.com)

Claude API pricing

Currency: USD. Unit: per 1 million tokens. The following applies to Anthropic’s first-party Claude API, using global inference, as checked on November 16, 2026. (platform.claude.com)

Tier Input Output Conditions
Standard $5 $25 Applies across the full 1-million-token context window.
Batch API $2.50 $12.50 Asynchronous batch processing; 50% discount on standard input and output.

For standard global inference, prompt-cache writes cost $6.25 per million input tokens for a five-minute cache or $10 per million input tokens for a one-hour cache; cache reads cost $0.50 per million input tokens. Specifying US-only inference adds a 10% multiplier to token pricing, including cache pricing. Server-side tools can incur separate charges. (platform.claude.com)

Official sources