Cursor's Composer 2.5 matches Opus 4.7 and GPT-5.5 benchmarks at a fraction of the cost

2026-05-18 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Cursor has released Composer 2.5, a significant update to its in-house AI coding model, building on Moonshot's open-source Kimi K2.5 checkpoint. This new iteration was trained with 25 times more synthetic tasks than its predecessor, Composer 2, allocating 85 percent of its compute budget to additional training and reinforcement learning. Composer 2.5 achieves benchmark scores of 79.8 percent on SWE-Bench Multilingual and 63.2 percent on CursorBench v3.1, matching the performance of Opus 4.7 and GPT-5.5. Notably, its pricing is substantially lower, at $0.50 per million input tokens and $2.50 per million output tokens, with a faster variant available for $3.00 and $15.00, respectively. Cursor is also developing a larger successor model "from scratch" with SpaceX and xAI, utilizing ten times the compute on the Colossus-2 cluster.

Key takeaway

For AI architects and engineering leaders evaluating coding assistants, Composer 2.5 offers a compelling alternative to higher-priced models like Opus 4.7 and GPT-5.5. Its comparable benchmark performance at significantly lower token costs could optimize your development budget without sacrificing code quality. Consider integrating Composer 2.5 into your workflows to reduce operational expenses.

Key insights

Composer 2.5 matches top-tier coding AI performance at a fraction of the cost through extensive synthetic task training.

Principles

• Synthetic data scales model capabilities.
• Cost-efficiency is achievable with competitive performance.

Method

Composer 2.5's development involved building on an open-source checkpoint, training on 25x more synthetic tasks, and dedicating 85% of compute to reinforcement learning.

In practice

• Evaluate Composer 2.5 for cost-sensitive coding tasks.
• Consider synthetic data generation for model improvement.

Topics

• Composer 2.5
• AI Coding Model
• Model Benchmarks
• Token Pricing
• Kimi K2.5

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• 100+ Tips: Cursor Composer 1.5 vs Claude Code Opus 4.6 vs Codex GPT 5.3 Spark + 2 Underdogs That Change the Math — Developers often overpay or under-utilize agentic coding tools due to infrequent re-evaluation.
• Introducing Composer 2.5 — Composer 2.5 improves AI assistant performance through targeted RL, synthetic data, and advanced distributed training.
• Cursor Composer 2.5: The AI coding agent that punches way above it’s price tag — High-performance AI coding agents can be achieved at significantly lower costs by leveraging existing powerful open-source models.
• GPT 5.5 vs Claude Opus 4.7 — Benchmarks offer controlled performance metrics, but real-world AI effectiveness depends on handling unstructured, unclear tasks.
• Composer 2: Cursor’s Bet That Specialized Models Beat General Ones at Their Own Game — Specialized AI models can outperform general models for specific tasks like coding, offering superior performance and cost.
• Cursor Composer 2.5 Is REALLY Good & On Par With Opus 4.7 & GPT 5.5? (Fully Tested) — Composer 2.5 offers a high speed-to-intelligence ratio, making it a cost-effective coding agent despite some quality gaps in web design.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.