Cursor Composer 2.5

Overview

Cursor Composer 2.5 is Cursor’s in-house coding model, launched in May 2026 as the successor to Composer 2. The model reports parity with Claude Opus 4.7 and GPT-5.5 on Cursor’s own coding benchmarks at a fraction of the per-token cost, extending Cursor’s in-house-model-plus-frontier-API arc since 2026-03-21-AI-Digest.

Timeline

2026-05-19-AI-Digest — Cursor launches Composer 2.5, reporting SWE-Bench Multilingual at 79.8% and CursorBench v3.1 at 63.2% — drawing level with Claude Opus 4.7 and GPT-5.5 on the IDE’s own coding benchmarks. Pricing: $0.50 / $2.50 per million input / output tokens on the standard variant, with a faster $3 / $15 tier. Cursor’s framing puts a typical agentic task at under $1 versus up to $11 on a frontier-lab API. Independent public benchmark replication is the open test — Cursor’s benchmarks are run on its own infrastructure.

Key Developments

Benchmark Parity at Sub-$1 Per Task: SWE-Bench Multilingual at 79.8% and CursorBench v3.1 at 63.2% claim parity with frontier-lab flagships (Claude Opus 4.7, GPT-5.5) on the IDE’s own benchmarks. Whether this survives public independent benchmarking is the test to watch.
Pricing Pressure on Frontier APIs: The $0.50 / $2.50 standard pricing and $3 / $15 faster tier — paired with the “under $1 vs up to $11” framing for a typical agentic task — is the kind of cost line that pulls budget out of frontier-lab APIs for routine agentic work.
In-House-Model-Plus-Frontier-API Arc: Composer 2.5 extends the strategy Cursor has been running since Composer 2 in 2026-03-21-AI-Digest — operate its own model for the high-volume routine path while keeping frontier-lab APIs available for harder tasks.

Cursor Composer 2.5

Overview

Timeline

Key Developments

See Also