Cursor's New Coding Model Matches Opus Performance at Fraction of Cost
TL;DR · AI Summary
Cursor's self-developed coding model Composer 2.5 matches Opus 4.7 performance with less than 1-point score gap, but costs 10-30x less: input tokens drop from $15 to ~$1.5 per million, output from $75 to ~$2.5. It shows clear improvements over Composer 2 in long-context tasks, complex instruction following, and collaboration smoothness.
Key Takeaways
- Composer 2.5 scores within 1 point of Opus 4.7, placing them in the same perform
- Significant cost advantage: 10x cheaper input and 30x cheaper output compared to
- Clear improvements in long-context task handling, complex instruction compliance
Outline
Jump quickly between sections.
Cursor releases self-developed coding model Composer 2.5, claiming performance comparable to Opus 4.7
Composer 2.5 scores within the same range as Opus 4.7, with maximum gap under 1 point
Composer 2.5 costs one-tenth of Opus 4.7 for input and one-thirtieth for output
Shows significant improvements over Composer 2 in long-task execution, complex instruction following, and collaboration smoothness
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- Cursor Composer 2.5模型发布
- 性能对标
- 评分差距<1分
- 与Opus 4.7同区间
- 价格优势
- 输入便宜10倍
- 输出便宜30倍
- 能力提升
- 长任务推进
- 复杂指令遵循
- 协作顺滑度
Highlights
Key sentences worth saving and sharing.
Composer 2.5 scores within the same range as Opus 4.7, with maximum gap under 1 point
Opus 4.7 costs ~$15 per million input tokens and $75 for output, while Composer 2.5 is 10x cheaper for input and 30x cheaper for output
Long tasks can continuously progress across rollout spans of hundreds of thousands of tokens without easily deviating
Complex instruction following is more reliable, with more stable calibration of communication style and engagement level
# Xiaohu on X: "Performance comparable to Opus, but 30 times cheaper? Cursor launches its self-developed coding model Composer 2.5
In terms of scoring: All results of Composer 2.5 fall within the same range as Opus 4.7, with the maximum difference being less than 1 point.
Price-wise: Opus 4.7 costs approximately $15 per million input tokens and $75 per million output tokens. Composer 2.5 is 10 times cheaper for input and 30 times cheaper for output.
Cursor claims Composer 2.5" / X
Don’t miss what’s happening
[](https://x.com/xiaohu)
[Xiaohu](https://x.com/xiaohu)
[@xiaohu](https://x.com/xiaohu)
Show translation
Performance comparable to Opus, but 30 times cheaper? Cursor launches its self-developed coding model Composer 2.5. In terms of scoring: All results of Composer 2.5 fall within the same range as Opus 4.7, with the maximum difference being less than 1 point. Price-wise: Opus 4.7 costs approximately $15 per million input tokens and $75 per million output tokens. Composer 2.5 is 10 times cheaper for input and 30 times cheaper for output. Cursor claims Composer 2.5 shows significant improvements over Composer 2 in both intelligence and behavioral performance, especially in long-duration tasks, complex instruction following, and collaborative smoothness. Long tasks can progress consistently through rollouts spanning hundreds of thousands of tokens without easily deviating. Complex instruction following is more reliable, with more stable calibration of communication style and engagement level, and enhanced work intensity calibration.
[](https://x.com/xiaohu/status/2056556014254825776/photo/1)
[2:02 AM · May 19, 2026](https://x.com/xiaohu/status/2056556014254825776)
·
[36.3K Views](https://x.com/xiaohu/status/2056556014254825776/analytics)
38
10
71
38
Read 38 replies