Anthropic released Claude Opus 4.7 as a general-availability model on April 16, 2026, framing it as a step up over Opus 4.6 in software engineering, complex agentic work, and multimodal understanding.
What's new
- Coding — measurable jump on real-world software engineering benchmarks (numbers below).
- Vision — supports images up to 2,576 pixels on the long edge, with gains in instruction following on multimodal tasks.
- Agents — Anthropic claims state-of-the-art performance on finance-agent and knowledge-work evaluations (GDPval-AA).
The benchmark numbers
The headline result is software engineering. On SWE-bench Pro — the harder, contamination-resistant version of SWE-bench — Opus 4.7 leads every shipping competitor.
Coding
| Benchmark | Opus 4.7 | Opus 4.6 | GPT-5.4 | Gemini 3.1 Pro |
|---|---|---|---|---|
| SWE-bench Verified | 87.6% | 80.8% | — | 80.6% |
| SWE-bench Pro | 64.3% | 53.4% | 57.7% | 54.2% |
| Terminal-Bench 2.0 | 69.4% | 65.4% | 75.1% ¹ | 68.5% |
Agents & tool use
| Benchmark | Opus 4.7 | Opus 4.6 | GPT-5.4 Pro | Gemini 3.1 Pro |
|---|---|---|---|---|
| MCP-Atlas (tool orchestration) | 77.3% | 75.8% | 68.1% | 73.9% |
| Finance Agent v1.1 | 64.4% | 60.1% | 61.5% | 59.7% |
| OSWorld-Verified (computer use) | 78.0% | 72.7% | 75.0% | — |
| BrowseComp (agentic search) | 79.3% | 83.7% | 89.3% | 85.9% |
Reasoning & knowledge
| Benchmark | Opus 4.7 | Opus 4.6 | GPT-5.4 Pro | Gemini 3.1 Pro |
|---|---|---|---|---|
| GPQA Diamond | 94.2% | 91.3% | 94.4% | 94.3% |
| Humanity's Last Exam (no tools) | 46.9% | 40.0% | 42.7% | 44.4% |
| Humanity's Last Exam (with tools) | 54.7% | 53.3% | 58.7% | 51.4% |
Vision & multilingual
| Benchmark | Opus 4.7 | Opus 4.6 | Gemini 3.1 Pro |
|---|---|---|---|
| CharXiv Reasoning (no tools) | 82.1% | 69.1% | — |
| CharXiv Reasoning (with tools) | 91.0% | 84.7% | — |
| MMMLU (multilingual) | 91.5% | 91.1% | 92.6% |
¹ GPT-5.4 used a self-reported harness, not directly comparable.
Other published results
- CursorBench — 70% (Opus 4.6: 58%)
- BigLaw Bench (Harvey) — 90.9% accuracy at high effort
- OfficeQA Pro (Databricks) — 21% fewer errors than Opus 4.6
- Rakuten-SWE-Bench — 3× more production tasks resolved than Opus 4.6
- GDPval-AA — Anthropic claims state-of-the-art (no specific score published)
The pattern: Opus 4.7 owns real-world software engineering (SWE-bench Verified and Pro), wins most agentic and tool-use evals outside of search, ties or narrowly trails GPT-5.4 Pro on graduate-level reasoning, and posts the largest jump anywhere on chart and document understanding (CharXiv).
Pricing and availability
Pricing is unchanged from 4.6: $5 per million input tokens and $25 per million output tokens. Opus 4.7 is available across Claude.ai, the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Why it matters for working professionals
For lawyers, researchers, and analysts already using Claude Projects (in Nowrap's tools directory), the upgrade is automatic on Pro and Team plans. Where it shows up most: longer documents, harder reasoning, and tighter agentic loops that previously needed manual steering.
Sources: anthropic.com/news/claude-opus-4-7 · benchmark breakdown via Vellum · competitor scoring via TheNextWeb