← Back
Anthropic
Anthropic releases 1M token context window for Claude Opus 4.6 and Sonnet 4.6 at standard pricing
Claude · releaseapifeaturemodel · platform.claude.com ↗

1M Token Context Window Now GA

Anthropic has expanded access to its 1M token context window, moving it from beta to general availability for Claude Opus 4.6 and Sonnet 4.6. Developers can now use requests exceeding 200k tokens without requiring a beta header, with long-context pricing automatically applied to requests over 200k input tokens.

Key Updates

Context Window & Rate Limits

  • The 1M context window is now GA for Opus 4.6 and Sonnet 4.6 at standard pricing
  • Requests over 200k tokens work automatically with no beta header required
  • Unified account limits now apply across all context lengths (dedicated 1M rate limits have been removed)
  • Media limits increased from 100 to 600 images or PDF pages per request when using 1M context window

Extended Thinking Display Control Developers can now omit thinking content from responses for faster streaming by setting thinking.display: "omitted". This preserves thinking block signatures for multi-turn continuity while hiding the thinking content itself. Billing remains unchanged.

Automatic Prompt Caching The Messages API now supports automatic prompt caching. Add a single cache_control field to your request body and the system automatically caches the last cacheable block, moving the cache point forward as conversations grow. This eliminates manual breakpoint management and works alongside existing block-level cache control for fine-grained optimization.

Model Deprecations

  • Claude Sonnet 3.7 and Claude Haiku 3.5 have been retired; all requests now return an error
  • Claude Haiku 3 deprecation announced, with retirement scheduled for April 19, 2026
  • Migration recommended to Sonnet 4.6 and Haiku 4.5 respectively

Developer Action Items

  • Update API calls to use Opus 4.6 or Sonnet 4.6 for 1M token support without beta headers
  • Evaluate automatic caching for prompt optimization in Messages API
  • Migrate away from deprecated models (Sonnet 3.7, Haiku 3.5, Haiku 3) before April 19, 2026
  • Test extended thinking display controls with thinking.display: "omitted" for faster streaming