Detailed Analysis
Claude Opus 4.7, released April 16, 2026, represents Anthropic's most capable generally available model to date, positioning itself at the top of the competitive frontier AI landscape among publicly accessible systems. The model delivers measurable benchmark advances over its predecessor, Opus 4.6, including a score of 64.3% on SWE-Bench Pro — a gain of over ten percentage points — along with 87.6% on SWE-Bench Verified and 69.4% on Terminal-Bench 2.0. It also outperforms OpenAI's GPT-5.4 and Google's Gemini 3.1 Pro across agentic coding, multidisciplinary reasoning, scaled tool use, and agentic computer use. Notably, Anthropic maintains a tiered release structure: Opus 4.7 sits below Claude Mythos Preview, a more powerful model restricted to partners participating in the closed Project Glasswing program, signaling that the company continues to gate its most advanced capabilities behind selective access arrangements.
The technical improvements in Opus 4.7 reflect a deliberate focus on agentic and long-horizon use cases. Vision capabilities have been substantially upgraded, with high-resolution image support extending to 2,576 pixels (3.75 megapixels), enabling more precise performance on document understanding, chart analysis, and file formats such as .docx and .pptx with self-verification. Memory and autonomy improvements — particularly file-system-based memory for scratchpads and multi-session workflows — push the model toward greater utility in persistent, asynchronous task environments. Behavioral refinements include more precise instruction-following, reduced sycophancy, and adaptive effort scaling, where even low-effort Opus 4.7 outputs are said to approximate medium-effort Opus 4.6 responses. New API controls including an "extra high effort" parameter, beta task budgets, and an expanded auto mode for permission management further extend developer control over the model's operational envelope.
Pricing for Opus 4.7 holds steady at $5 per million input tokens and $25 per million output tokens — the same rate as its predecessors — though Anthropic notes that high-resolution image processing and maximum effort settings will increase token consumption in practice. The model is available across Anthropic's own API, web and desktop applications, Claude Code, Amazon Bedrock, and GitHub Copilot, where it is set to replace Opus 4.5 and 4.6 with a promotional 7.5× token multiplier through April 30, 2026. The "Ultra review" slash command for code analysis, available to Max subscribers through Claude Code, indicates continued investment in developer tooling as a primary growth vector for the Claude product line.
Opus 4.7's release reflects a broader industry trend in which frontier AI labs increasingly differentiate between their publicly released models and restricted, pre-release systems reserved for strategic partners. By maintaining Claude Mythos Preview as a closed offering under Project Glasswing while positioning Opus 4.7 as the "most capable generally available" model, Anthropic mirrors patterns seen across competitors — using tiered access to manage safety evaluation timelines, cultivate enterprise relationships, and control the pace at which the most powerful capabilities reach the broader market. The emphasis on agentic performance, memory persistence, and long-running workflows also reflects a maturing understanding within the industry that raw benchmark scores on static tasks are insufficient proxies for real-world utility; the competitive battleground is increasingly defined by how well models sustain coherent, goal-directed behavior across extended, multi-step environments.
The reception among developers, who describe Opus 4.7 as a meaningful upgrade for production-grade software engineering and enterprise automation, underscores the growing commercial stakes attached to agentic AI reliability. Anthropic's positioning of the model as a "better coworker" — one that requires fewer corrections and handles ambiguity more robustly — speaks to the company's long-standing emphasis on combining capability with predictable, trustworthy behavior. Whether the performance advantages over GPT-5.4 and Gemini 3.1 Pro hold up under sustained real-world deployment, and how quickly competitors respond, will determine whether Opus 4.7 consolidates Anthropic's standing or serves primarily as a temporary benchmark leader in an accelerating release cycle.
Read original article →