Cursor's Composer 2.5 Matches Claude Opus 4.7 on Coding Benchmarks at One-Tenth the Price

Cursor on May 18-20, 2026 released Composer 2.5, the third generation of its proprietary coding model, and is openly pitching it as a frontier-tier alternative to Claude Opus 4.7 at roughly a tenth of the per-token cost. The model is exclusive to the Cursor IDE and the @cursor/sdk — Cursor is not selling it as a general-purpose API — but inside that environment the pricing math is hard to ignore.

Standard pricing lands at /usr/bin/bash.50 per million input tokens and .50 per million output tokens, against /5 for Claude Opus 4.7. On benchmarks the gap is much narrower than the price gap: Composer 2.5 hits 79.8% on SWE-Bench Multilingual — effectively a tie with Opus 4.7's 80.5% — and 63.2% on CursorBench v3.1 at default effort, edging out Opus 4.7 at 61.6% and GPT-5.5 at 59.2%. On terminal tasks the two models are essentially indistinguishable.

The strategic message from Cursor is that vertically-integrated coding agents can now compete head-on with the frontier labs on quality while undercutting them on cost, because the IDE owner controls the entire context-gathering and tool-use loop. Anthropic and OpenAI both have to ship general-purpose models that perform well in any harness; Cursor only has to be excellent inside Cursor. That focus shows up in latency and tool-call accuracy as much as in the headline benchmark scores.

For developers, the immediate read is that the choice between Claude Code, OpenAI Codex, and Cursor is increasingly a price-performance question rather than a quality one. For Anthropic — whose enterprise revenue story leans heavily on Claude Code adoption — Composer 2.5 is the first credible signal that the labs no longer have a monopoly on top-tier coding intelligence. Expect token prices on Opus to drift down before the end of Q3.