Anthropic's newest flagship pairs a 1M-token context with sharper agentic reasoning — and a price drop that makes long-document workflows viable at scale.
Anthropic has released Claude Opus 4.7, a model that combines a one-million token context window with a sizeable jump in agentic coding ability. The release lands almost exactly six months after Opus 4.5 and continues Anthropic's pattern of shipping minor version bumps that, in practice, behave like generational leaps.
The headline number is context length. Where Opus 4.5 capped out at 200K tokens, 4.7 ingests a full million — enough to hold the entire Linux kernel source, a quarter of Wikipedia, or roughly 75 hours of meeting transcripts in a single prompt. Crucially, Anthropic also published needle-in-a-haystack evals showing recall stays above 98% across the full window, a result that has eluded earlier long-context models.
Where the gains show up
On SWE-Bench Verified, Opus 4.7 hits 79.4% — up from 73.1% on 4.5 and now the highest published score for any model. The improvement is concentrated in multi-file refactors and bug fixes that require holding a large codebase in working memory, which is precisely where the longer context pays off.
Pricing and availability
The 1M-token tier is priced at $6 per million input tokens and $30 per million output, with prompt-cache reads at $0.60. That is more expensive than Sonnet 4.6 but roughly half the per-token cost of running GPT-5 at comparable context length. Opus 4.7 is available today in the Claude API, on Amazon Bedrock, and on Google Cloud Vertex AI.