Anthropic · 2026-04-16 · seismic
Claude Opus 4.7 — 87.6% SWE-bench, 3.75MP Vision, Task Budgets, at Same $5/$25 Pricing
Anthropic shipped Claude Opus 4.7: 87.6% SWE-bench Verified, 64.3% SWE-bench Pro, 3.75MP vision (3× previous), 128k max output, xhigh effort level, and task budgets in beta. Same $5/$25 pricing as Opus 4.6; breaking API changes.
Anthropic's most capable publicly available model: double-digit coding gains, 3× vision resolution, and a new effort system — same price as the model it replaces.
Key specs
| Context window | 1M tokens |
|---|---|
| Price (input) | $5 / MTok |
| Price (output) | $25 / MTok |
| SWE-bench | 87.6% |
| Swe bench pro | 64.3% |
| Cursor bench | 70% (vs 58% Opus 4.6) |
| Rakuten swe bench | 3× more tasks vs Opus 4.6 |
| Visual acuity benchmark | 98.5% (vs 54.5% Opus 4.6) |
| Max image resolution | 2576px / 3.75MP |
| Big law bench | 90.9% |
| Max output tokens | 128k |
| Tokenizer overhead | 1.0–1.35× more tokens vs Opus 4.6 |
What is it?
Claude Opus 4.7 launched on April 16, 2026 as Anthropic's most capable generally available model. It scores 87.6% on SWE-bench Verified and 64.3% on SWE-bench Pro, resolves 3× more production tasks than Opus 4.6 on the Rakuten-SWE-Bench (enterprise codebase) benchmark, and scores 70% on CursorBench (vs 58% for 4.6). Vision gets a step-change: maximum input resolution jumps from 1568px to 2576px (3.75 megapixels, over 3× the pixels), with 1:1 coordinate mapping so computer-use and screenshot workflows no longer need scale-factor math. Context window is 1M tokens, max output is 128k tokens, and pricing is unchanged at $5/M input and $25/M output.
How does it work?
The model's intelligence gains come partly from a new tokenizer (producing 1.0–1.35× more tokens from the same text — see the migration guide before upgrading). Two new API features ship with Opus 4.7: task budgets (a beta feature that gives the model an advisory token target for an entire agentic loop, including thinking, tool calls, and output, so it can self-regulate on long horizon tasks) and a new xhigh effort level above the previous high ceiling. Both require the Messages API. Breaking changes: extended thinking budget tokens are removed (adaptive thinking only, off by default), and temperature, top_p, and top_k parameters will return 400 errors if set to non-default values.
Why does it matter?
The 4.7 tokenizer is a genuine hidden cost increase for text workloads: at the same $5/M token rate, expect up to 35% more input spend when migrating from Opus 4.6, though the model's intelligence improvements may reduce the total number of turns needed per task. For agentic coding, the 3× Rakuten-SWE-Bench improvement means Opus 4.7 handles the hardest class of enterprise codebase tasks that previously needed close human supervision. The task budget API gives operators a new lever for cost control on long-horizon workflows without sacrificing quality on short tasks.
Who is it for?
Teams running agentic coding pipelines on Claude API; developers doing vision-heavy work (computer use, document extraction, chart analysis); anyone migrating from Opus 4.6.
Try it
model='claude-opus-4-7' — check platform.claude.com/docs/en/about-claude/models/migration-guide before upgrading