Anthropic has made Claude Opus 4.7 generally available across its products and partners, describing it as an improvement over Opus 4.6 in advanced software engineering and high-resolution vision capabilities.
“Users report being able to hand off their hardest coding work—the kind that previously needed close supervision—to Opus 4.7 with confidence,” the company said in its announcement.
Opus 4.7 handles complex, long-running tasks with attention to instructions and verifies its own outputs. It also processes images at higher resolution, accepting inputs up to 2,576 pixels on the long edge, more than three times the capacity of prior models.
The model shows better results than Opus 4.6 on several benchmarks, though it remains less capable than Anthropic’s Claude Mythos Preview.
“Opus 4.7 has better vision for high-resolution images,” Anthropic said. “This opens up a wealth of multimodal uses that depend on fine visual detail.”
The model also improves memory across sessions using file system-based storage and performs better on finance analysis tasks, including state-of-the-art results on the Finance Agent evaluation and GDPval-AA benchmark.
Pricing stays at $5 per million input tokens and $25 per million output tokens. It is available through the Claude API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.
Anthropic applied safeguards to Opus 4.7 to block prohibited cybersecurity uses, following last week’s announcement of Project Glasswing.
Under Project Glasswing, Anthropic is providing selected organisations with early access to Mythos to help secure critical software systems and prepare industry-wide defences against emerging AI-driven cyber risks. Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks are a part of this initiative.
Mythos Preview remains the best-aligned model according to Anthropic’s evaluations.
The company tested new cyber safeguards on Opus 4.7 before considering broader release of Mythos-class models. “During its training we experimented with efforts to differentially reduce these capabilities,” Anthropic said.
Security professionals can join the new Cyber Verification Program for legitimate uses such as vulnerability research and penetration testing.
Early testers noted stronger instruction following, with the model now interpreting prompts more literally. This requires users to adjust existing prompts.
Alongside the release, Anthropic introduced an “xhigh” effort level for finer control over reasoning depth and latency. In Claude Code, the default effort level rises to xhigh for all plans.
Additional updates include task budgets in the API, now in public beta, and a new /ultrareview command in Claude Code. Auto mode, which allows Claude to make decisions on behalf of users, has also been extended to Max plan users.
Anthropic said the model uses an updated tokeniser, which may increase token counts by 1.0–1.35× for the same input. It added that higher effort settings may also increase output tokens, though internal testing showed overall efficiency gains in coding tasks.
On safety, the company said Opus 4.7 shows a similar profile to Opus 4.6, with improvements in honesty and resistance to prompt injection attacks on some measures.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.