Claude Opus 4.7

// Overview

Claude Opus 4.7 is the Opus-tier flagship Anthropic released on April 16, 2026, succeeding Claude Opus 4.6 and later superseded by Claude Opus 4.8. It became generally available across Claude.ai, the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry, and is positioned for the hardest reasoning, coding, and long-horizon agentic work.

Claude Opus 4.7 ships with a 1M-token context window at standard pricing and can generate up to 128K output tokens. It accepts text and image input and uses adaptive thinking only — sampling parameters (temperature, top_p, top_k) and the budget_tokens control were removed in this release. Opus 4.7 also added the xhigh effort level, sitting between high and max for finer control over the reasoning-versus-latency tradeoff.

Versus Opus 4.6, Claude Opus 4.7 brought substantially stronger software engineering (87.6% on SWE-bench Verified, up from 80.8%), markedly improved vision (images up to 2,576px on the long edge, roughly 3.75 megapixels), and better instruction-following and agentic reliability. Pricing was unchanged at $5 per million input tokens and $25 per million output tokens.

// Benchmarks

1. SWE-bench Verified87.6%
2. SWE-bench Pro64.3%
3. Terminal-Bench 2.069.4%
4. GPQA Diamond94.2%
5. OSWorld-Verified (computer use)78%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Prompt-cache writes are $6.25 (5-min) and $10 (1-hour) per million tokens. Batch API is $2.50 input / $12.50 output. An optional research-preview Fast mode is $30 input / $150 output per million tokens.

Pricing source ↗

// Strengths

• Frontier software engineering — 87.6% on SWE-bench Verified and 64.3% on the harder SWE-bench Pro
• Long-horizon agentic coding and tool use across long-running tasks
• High-resolution vision up to ~3.75 MP (2,576px long edge) for screenshots, charts, and documents
• 1M-token context window billed at standard per-token rates with no long-context premium
• Adaptive thinking with five effort levels (low through max, including the new xhigh)
• Strong graduate-level reasoning — 94.2% on GPQA Diamond

// Best for

• Reach for it for multi-step agentic coding and complex, long-running software-engineering tasks across a large codebase.
• Use it when a task needs deep reasoning over very long documents within a 1M-token context.
• Apply it to vision-heavy work — computer use, screenshot and chart understanding, and document analysis — that benefits from high-resolution image input.
• Choose it for high-autonomy agent workflows where reliability and instruction-following matter more than per-token cost.

// How to access

// Claude Opus — every version

The full lineage of the Claude Opus line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.