Cursor ships Composer 2, its second in-house coding model

Anysphere (Cursor)/Cursor

Cursor released Composer 2, built on continued pretraining followed by reinforcement learning on long coding tasks. The model scores 61.3 on CursorBench versus 44.2 for Composer 1.5 and 73.7 on SWE-bench Multilingual. Pricing: $0.50/M input and $2.50/M output tokens for the standard variant, with a faster variant at $1.50/M and $7.50/M.

• CursorBench 61.3 vs Composer 1.5's 44.2 and Composer 1's 38.0; Terminal-Bench 2.0: 61.7 vs prior 47.9; SWE-bench Multilingual: 73.7 vs 65.9
• Trained with continued pretraining and reinforcement learning on long-horizon coding tasks requiring hundreds of actions
• Standard pricing: $0.50/M input, $2.50/M output; faster variant: $1.50/M input, $7.50/M output with same intelligence
• Frontier-level at coding and solving challenging tasks; faster variant offers lower costs versus other fast models