Anthropic released Claude Opus 4.7 on April 16. It is their most capable commercial model right now, with real improvements in coding, reasoning, and agentic tasks. It is also the first Opus release where Anthropic openly concedes in the announcement that it is not their best model. That is worth understanding before you decide what to do with it.

What Is Actually Better

Opus 4.7 outperforms Opus 4.6 across the areas that matter most for developer and agentic workloads:

  • SWE-bench Pro: 64.3% — a real number on a real benchmark, not a cherry-picked internal test
  • Internal coding benchmark: 13% improvement over Opus 4.6 on a 93-task coding suite
  • Agentic computer use, multidisciplinary reasoning, and scaled tool use all improved over 4.6

The model also has better vision capabilities and a new ability to check its own work before returning an answer. That self-verification step is useful for agentic tasks where small errors early in a chain compound into bigger problems downstream.

The Mythos Situation

Anthropic was unusually direct in the Opus 4.7 announcement. They said, in plain language, that Opus 4.7 is "less broadly capable than our most powerful model, Claude Mythos Preview."

Mythos is the frontier model Anthropic revealed earlier this month under Project Glasswing. It is not publicly available. Around 40 vetted enterprise and government partners have access to it. Everyone else is working with Opus 4.7.

This is different from how AI releases usually work. Companies generally announce what a model can do, not what it falls short of. Anthropic leading with that caveat suggests they are either being genuinely transparent, or they are managing expectations ahead of a broader Mythos release that has not happened yet. Probably both.

For most businesses, Mythos is not relevant right now. You cannot get access to it unless you are already in that partner group. Opus 4.7 is what you are working with.

Pricing

Same as Opus 4.6: $5 per million input tokens, $25 per million output tokens. No price increase with the new release, which is notable given the performance improvements.

Where It Is Available

Opus 4.7 is live across all Claude.ai plans, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. If you are already using Opus 4.6 through any of these platforms, you can switch now. There is no waitlist.

Should You Switch?

If you are doing serious coding work, agentic tasks, or anything that involves long chains of reasoning, yes. The SWE-bench improvement is not trivial, and the self-verification capability is useful for reducing errors in automated workflows.

If you are using Claude for writing, research, or lighter tasks where Sonnet 4.6 already does the job, this release probably does not change your setup. Sonnet remains the better cost-efficiency option for most everyday workloads.

The Mythos situation is worth watching. Anthropic has a frontier model behind a partner program, and Opus 4.7 is the public ceiling for now. That ceiling is still high enough for most things businesses actually need to do.