Anthropic ships Claude Opus 4.7 with 1M-token context window
Anthropic released Opus 4.7 yesterday with a full 1M-token context window, sharper agent loop reliability, and faster tool-use latency. Replaces Opus 4.5 as the recommended default for long-context coding and multi-step workflows.
Anthropic released Claude Opus 4.7 on May 14, 2026. The headline change is the context window: 1M tokens, up from 200K in 4.6. Practically, that puts entire codebases, full RFC threads, and hour-long meeting transcripts within the routine input budget.
What changed since 4.6
Three deltas worth tracking:
- Context window: 1M tokens (5× the previous limit).
- Tool-use latency: noticeably faster on multi-tool agentic loops in our early testing.
- Agent loop reliability: fewer “model stops mid-task” cases, especially on long autonomous runs.
Where it lands in the cohort
GPT-5 remains the broader default for cost-sensitive general workloads. Gemini 2.5 Pro still leads on video and long-document understanding. Opus 4.7’s lane is agentic reliability and long-context coding — that’s the bet Anthropic is making, and on those tasks the new release widens the gap.
Practical impact
If you’re running Claude Code subscriptions, Opus 4.7 is the default model on the included tier. API users should see Opus 4.7 in the model selector immediately; pricing parity with Opus 4.6 has not been confirmed at publish time.
We’ll publish a 30-day follow-up scorecard on June 13 with usage data and a deeper look at whether the agent-reliability claims hold up.
- Anthropic announcement www.anthropic.com
- Anthropic API docs — Opus 4.7 docs.claude.com