Hand off your hardest code: Claude Opus 4.7 is here!

Hey everyone, it's Shii-chan! Today we get to talk about a brand-new model, and I've been buzzing about it since morning!

Anthropic News anthropic.com

What was announced?

Over on Anthropic's News, they announced that a new flagship model, Claude Opus 4.7, is now generally available! It went GA on April 16, 2026.

In short, it's a solid step up from the previous model, Opus 4.6 - especially on hard software engineering. Users report they can now hand off their hardest coding work, the kind that used to need close supervision, with confidence. It handles long, complex tasks with rigor, follows instructions precisely, and even verifies its own outputs before reporting back. So diligent!

And while it's less broadly capable than Anthropic's most powerful model, Claude Mythos Preview, it beats Opus 4.6 across a range of benchmarks.

The story so far

Opus 4.7 is a direct upgrade to Opus 4.6, so it's less "rip and replace" and more "swap in and get stronger."

For context, last week Anthropic announced Project Glasswing, talking about both the risks and the benefits of AI for cybersecurity. The plan was to keep Mythos Preview's release limited and test new cyber safeguards on less capable models first. Opus 4.7 is the first of those: during training they deliberately held its cyber capabilities below Mythos Preview, and it ships with safeguards that automatically detect and block prohibited or high-risk cybersecurity requests.

Security professionals doing legitimate work - vulnerability research, penetration testing, red-teaming - can apply for the new Cyber Verification Program.

What changes

The biggest wins are in coding and long-running agentic tasks. Instruction-following got much tighter, and it handles async workflows like CI/CD and automation well.

Vision got a big upgrade too. It can now take images up to 2,576 pixels on the long edge (~3.75 megapixels), more than three times prior Claude models! That helps computer-use agents reading dense screenshots, data extraction from complex diagrams, and reading chemical structures and technical diagrams.

On top of that, it's more tasteful on professional work like slides, docs, and UIs, and its file-system-based memory improved, so it remembers key notes across long, multi-session work.

Dive Deep

Pricing is unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens (pricing). It's available across Claude products and the API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. Developers can use claude-opus-4-7 via the Claude API.

There's also a new effort option: xhigh (extra high) sits between high and max, giving finer control over the reasoning-vs-latency tradeoff on hard problems (effort docs). In Claude Code, the default effort was raised to xhigh on all plans, and the team recommends starting with high or xhigh for coding.

On the Claude Platform (API), task budgets launched in public beta to help guide token spend across longer runs, and there's higher-resolution image support (vision docs).

In Claude Code, the new /ultrareview slash command runs a dedicated review session that flags bugs and design issues a careful reviewer would catch. Pro and Max users get three free ultrareviews to try. Auto mode was also extended to Max users.

Two things to plan for when migrating: Opus 4.7 uses an updated tokenizer, so the same input can map to roughly 1.0-1.35x more tokens depending on content, and it thinks more at higher effort, producing more output tokens. Anthropic says the net effect was favorable in their coding evals. See the migration guide, and the system card for the safety details.

Overall the safety profile is similar to Opus 4.6, with low rates of deception and sycophancy, and some improvements like resistance to prompt injection.

Wrap-up

• Anthropic's News announced GA of the new flagship, Claude Opus 4.7
• Big gains in coding and long-running agentic work, plus tighter instruction-following
• Vision now supports up to 2,576 pixels on the long edge (~3.75 megapixels), 3x+ prior models
• Pricing unchanged ($5 input / $25 output per million tokens); use claude-opus-4-7
• New xhigh effort, task budgets (beta), /ultrareview, and expanded auto mode all launched alongside
• Cyber capabilities deliberately dialed down, with automatic safeguards

If you want to hand off your hardest code, you're running agents in parallel, or you work with high-resolution multimodal inputs, this is the release for you!