Claude Opus 4.7

Claude Opus 4.7 is Anthropic's most advanced publicly available language model, released on April 16, 2026. It is the direct successor to Claude Opus 4.6 (released February 2026) and belongs to the Claude 4 model family. The API model identifier is claude-opus-4-7. According to Anthropic, the model is less capable than the proprietary Claude Mythos Preview, but remains the strongest generally available model in the company's lineup.

Key improvements over Opus 4.6

Opus 4.7 delivers significant improvements across four areas. In agentic coding, the model achieved 87.6% on SWE-bench Verified (up from 80.8%) and 64.3% on SWE-bench Pro (up from 53.4%), setting a new benchmark among publicly available models. In vision, this is the first Claude model to support high-resolution images — up to 2576 pixels on the long side (~3.75 MP), compared to the previous limit of 1568 px / 1.15 MP. Image coordinates now map 1:1 to actual pixels, simplifying work in computer use mode.

A new effort level, xhigh (extra high) — positioned between high and max — has been introduced, allowing finer control over the trade-off between reasoning depth and response latency. Task budgets are available in public beta: a mechanism that lets developers set an approximate token limit for an entire agentic cycle, with a visible countdown counter accessible to the model.

The model substantially improves literal instruction-following — it interprets instructions literally rather than loosely, which requires adjusting prompts optimized for Opus 4.6. Opus 4.7 verifies its own outputs before responding, makes better use of filesystem-based memory, and more reliably executes long-running multi-session tasks. Adaptive thinking is supported and is the only thinking mode available — fixed extended thinking budgets have been removed.

API changes and cybersecurity safeguards

Opus 4.7 introduces breaking changes to the Messages API: (1) setting a fixed thinking budget via budget_tokens returns a 400 error; (2) setting a custom value for temperature, top_p, or top_k returns a 400 error; (3) thinking content is omitted from responses by default. The new tokenizer may produce up to 35% more tokens for the same input text compared to Opus 4.6. Opus 4.7 is the first Claude model on which Anthropic is testing automated cybersecurity safeguards ahead of a potential broader release of Mythos-class models. Security professionals conducting legitimate tasks (penetration testing, red-teaming, vulnerability research) may apply for access through the Cyber Verification Program.

Safety and model alignment

Anthropic's alignment evaluation found that Opus 4.7 is "largely well-aligned and trustworthy, though not fully ideal in its behavior." The model exhibits a safety profile similar to Opus 4.6, with improvements in honesty and resistance to prompt injection attacks. A minor regression was noted in one area: the model may provide overly detailed harm-reduction advice regarding controlled substances. The full safety evaluation is published in the Claude Opus 4.7 system card (232 pages, published April 16, 2026).