Anthropic shipped Claude Opus 4.7 today with the 1M token context window unlocked at the flagship tier. Previously this was a Sonnet-only feature gated behind beta access — Opus users had been stuck at 200K despite paying premium rates.
The release notes call out specific gains on SWE-bench Verified (+3.2 points over 4.6), Terminal-Bench, and a new internal benchmark for multi-day agent runs. Pricing stays at $15/$75 per million for input/output but jumps to $30/$150 above 200K tokens — same tier structure as Sonnet's long-context mode.
For Chinese-speaking developers the practical question is whether Opus 4.7 finally beats Sonnet 4.5 for agentic coding given the latter's lower cost. Early community testing suggests yes for tasks involving large codebases, no for routine refactors. Worth a structured eval before swapping in your CI pipeline.