Claude Opus 4.7

Anthropic released Claude Opus 4.7 on April 16, 2026, positioning it as a notable improvement over Opus 4.6 in advanced software engineering, particularly on the most difficult workloads. The model is generally available across Claude surfaces, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on day one. Opus 4.7 brings substantially higher-resolution vision, more polished output for professional tasks such as interfaces and slides, and self-verification behavior on long-running work. Anthropic ships cyber-use safeguards and a Cyber Verification Program, and pricing is unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens.

A Substack newsletter from Linas Beliūnas, dated June 1, 2026, catalogs 29 Anthropic product updates shipped in the first five months of 2026, including new Claude models, agents, desktop apps, a certification program, and multibillion-dollar funding rounds. It reports Anthropic's run-rate revenue rising from $14 billi As a third-party newsletter commentary rather than a first-party Anthropic release page, this article offers broad trend context for the Claude family in 2026 rather than version-specific technical detail. Its value is in the aggregate product-narrative framing — Claude's shift toward long-running agentic execution and

Claude Opus 4.8, Anthropic's latest Opus model, is now available in GitHub Copilot. In our early testing, Opus 4.8 demonstrates a clear step ...

A third-party timeline article on hidekazu-konishi.com, first published May 16, 2026 and last updated September 3, 2026, builds a complete release timeline for Anthropic's Claude foundation model family from Claude 1 in March 2023 through Claude Fable 5, Claude Sonnet 5 (June 2026), and Claude Opus 5 (July 2026). It do As a comprehensive family-level compilation rather than first-party Anthropic material, the page is useful for situating Claude Opus 4.7 within the broader Claude release chronology but does not provide version-specific technical detail about Opus 4.7 itself. The article is part of the author's history-and-timeline ser

Claude Opus 4.7, Anthropic's latest Opus model, is now rolling out on GitHub Copilot. In our early testing, Opus 4.7 delivers stronger multi ...

A third-party benchmark breakdown of Claude Opus 4.7 highlights SWE-bench Verified climbing to 87.6% and SWE-bench Pro reaching 64.3%, leapfrogging GPT-5.4 and Gemini 3.1 Pro on agentic engineering tasks. OSWorld-Verified rises to 78.0% alongside a 3x resolution boost, and MCP-Atlas leads at 77.3%, reinforcing Opus 4.7 as a strong tool-use model. The same analysis notes GPQA Diamond at 94.2% and a BrowseComp decline to 79.3%, signaling that agentic web search is the one area where Opus 4.7 trails peers. Bedrock availability and the $5/$25 per million token pricing parity with Opus 4.6 are reconfirmed, framing Opus 4.7 as a focused upgrade for production coding and tool-calling agents.

Anthropic's Claude Opus 4.7 is positioned as its most capable generally available model, with 1M-token context, up to 128K output tokens, and pricing held flat at $5 per million input tokens and $25 per million output tokens — same as Opus 4.6. The reliable knowledge cutoff is January 2026. The release is framed around Migration from Opus 4.6 carries two breaking API changes that affect Azure-hosted deployments: (1) setting non-default values for temperature, top_p, or top_k is expected to return a 400 error, so those sampling parameters must be removed and behavior steered entirely via prompting; and (2) explicit thinking budgets (t

Anthropic launched Claude Opus 4.7 on April 16, 2026, described as the company's most capable generally available model at that time and purpose-built for coding, long-horizon agentic tasks, and professional workflows. The release delivers significant improvements over Opus 4.6 in benchmark performance, vision quality, Reported benchmark gains, attributed to AWS launch coverage, include SWE-bench Pro rising to 64.3% from 53.4% (+10.9 points), SWE-bench Verified reaching 87.6% from 80.8% (+6.8 points), and Terminal-Bench 2.0 climbing to 69.4% from 65.4% (+4.0 points). The piece emphasizes that pushing past 87% on SWE-bench Verified pu