Anthropic made Claude Opus 4.7 generally available on April 16, 2026. The company says Opus 4.7 represents a meaningful upgrade over Opus 4.6 for advanced software engineering tasks, especially the hardest, long‑running, multi‑step workflows.
What Opus 4.7 improves
-
Coding and long‑context tasks: According to Anthropic and early testers, Opus 4.7 requires less close supervision on difficult coding work, follows instructions more precisely, performs internal checks, and verifies outputs before returning them. On a 93‑task coding benchmark, Opus 4.7 improved resolution by 13% compared to Opus 4.6, including solving four tasks that neither Opus 4.6 nor Sonnet 4.6 could solve.
-
Multimodal vision: The model accepts higher‑resolution images, with a maximum long edge of 2,576 pixels (~3.75 megapixels), more than three times the resolution supported by prior Claude models. This enables use cases requiring fine visual detail, such as dense screenshot parsing and complex diagram extraction.
-
Professional outputs and creativity: Anthropic reports that Opus 4.7 produces more tasteful and creative outputs for professional tasks like interfaces, slides, and documents.
Security and built‑in cyber safeguards
Opus 4.7 is released with automatic safeguards that detect and block requests indicative of prohibited or high‑risk cybersecurity uses. This follows Anthropic’s Project Glasswing approach of testing new cyber safeguards first on less capable models before broader Mythos‑class releases. Anthropic also opened a Cyber Verification Program for security professionals who want to use Opus 4.7 for legitimate security research (vulnerability research, penetration testing, red‑teaming).
Availability, pricing and integrations
Claude Opus 4.7 is available today across all Claude products and the API, and through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Pricing remains the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens. Developers can call claude‑opus‑4‑7 via the Claude API.
Early testing feedback and benchmark highlights
Anthropic shared results and customer feedback from early access testing:
-
Multiple customers reported improved performance on asynchronous workflows, automation, CI/CD, and long‑running tasks, with better error resilience, lower tool‑error variance, and more consistent execution.
-
Specific reported results include a 13% lift on a 93‑task coding benchmark. In an internal research‑agent benchmark, Opus 4.7 tied for top overall score across six modules at 0.715 and delivered the most consistent long‑context performance tested. On the General Finance module it scored 0.813 versus Opus 4.6’s 0.767. On CursorBench Opus 4.7 scored 70% compared with 58% for Opus 4.6. Rakuten‑SWE‑Bench reported Opus 4.7 resolved three times more production tasks than Opus 4.6, with double‑digit improvements in code and test quality.
-
Customer anecdotes and metrics: XBOW measured 98.5% on their visual‑acuity benchmark with Opus 4.7 versus 54.5% for Opus 4.6. Databricks’ OfficeQA Pro showed 21% fewer errors when Opus 4.7 worked with source information compared with Opus 4.6. Several companies (Replit, Notion Agent, Databricks, Vercel and others) reported that Opus 4.7 produced higher‑quality work more efficiently in their production or evaluation workloads.
Migration considerations
Anthropic notes two migration effects when moving from Opus 4.6 to Opus 4.7:
- Updated tokenizer: the same input can map to more tokens (roughly 1.0–1.35× depending on content).
- Higher effort thinking: at higher effort levels the model does more internal processing and can produce more output tokens, improving reliability on hard problems but increasing token use in some cases.
Users can manage token usage via the effort parameter, task budgets, or by prompting the model for concision. Anthropic’s internal tests showed a favorable net effect on token usage for coding evaluations, but they recommend measuring changes on real traffic. A migration guide is provided.
Additional features launched with Opus 4.7
-
New effort level: xhigh (extra high) sits between high and max to give finer control of the reasoning vs latency tradeoff. Claude Code sets xhigh as the default effort for applicable plans.
-
API task budgets (public beta): developers can guide Claude’s token spend across longer runs.
-
Claude Code updates: a new /ultrareview slash command creates a dedicated review session to flag bugs and design issues; Pro and Max Claude Code users receive three free ultrareviews. Auto mode (where Claude makes certain decisions autonomously) has been extended to Max users.
Safety, alignment and limitations
Anthropic reports that Opus 4.7’s safety profile is broadly similar to Opus 4.6: automated audits show low rates of concerning behaviors such as deception, sycophancy, and cooperation with misuse. On some measures (honesty and resistance to malicious prompt injection) Opus 4.7 improves on Opus 4.6; on others (for example a tendency to give overly detailed harm‑reduction advice on controlled substances) it is modestly weaker. Anthropic characterizes the model as “largely well‑aligned and trustworthy, though not fully ideal in its behavior.” Mythos Preview remains the best‑aligned model the company has trained according to internal evaluations. Full safety evaluations are included in the Claude Opus 4.7 System Card.
Document updates
The public documentation was updated on May 4, 2026 to reflect revised OfficeQA Pro scores for Opus 4.7 in the Document reasoning graph.
