According to The Rundown AI on X, Cursor’s in-house coding model Composer 2 Fast delivers output tokens at $7.50 per million compared with $75 for GPT-5.4 Fast and $150 for Opus 4.6 Fast, making it 10–20x cheaper to run (source: The Rundown AI). As reported by The Rundown AI, Terminal-Bench 2.0 scores are 61.7 for Composer 2, 58.0 for Opus 4.6, and 75.1 for GPT-5.4, indicating Composer 2 surpasses Anthropic’s Opus 4.6 while narrowing the gap with OpenAI’s GPT-5.4 (source: The Rundown AI). According to The Rundown AI, on CursorBench—Cursor’s internal evaluation built from real coding sessions—Composer 2 ranks just below GPT-5.4 and above Opus 4.6 at a fraction of the per-task cost, highlighting immediate opportunities to cut unit economics for code generation, code review, and refactoring workloads (source: The Rundown AI). For engineering leaders and platform teams, the business impact includes lower inference spend, expanded coverage for CI automation, and the ability to pilot multi-model routing where cost-sensitive tasks default to Composer 2 while complex tasks escalate to GPT-5.4 (source: The Rundown AI).