Composer 2.5 sits with the frontier coding agents
Cursor Composer 2.5 costs a fraction of the price of the very best models from the frontier labs, and Artificial Analysis puts it almost at the top of the pack in their aggregated coding agents benchmark. It comes 3rd overall, trailing only Claude Opus 4.7 Max and ChatGPT 5.5 X-high.
Artificial Analysis Coding Agent Index. Screenshot from 26 May 2026.
In other words, if the benchmark is to be believed, it’s a better coding tool than Opus 4.6, GPT 5.5 High, Claude Sonnet 4.6 on any thinking setting, etcetera. And it’s cheap.
| Model / AI System | Provider | Input Price (Per 1M Tokens) | Output Price (Per 1M Tokens) |
|---|---|---|---|
| Claude Opus 4.7 | Anthropic | $5.00 | $25.00 |
| ChatGPT 5.5 | OpenAI | $5.00 | $30.00 |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 |
| Cursor Composer 2.5 (Standard) | Cursor | $0.50 | $2.50 |
| Cursor Composer 2.5 (Fast) | Cursor | $3.00 | $15.00 |
Once more, if we measure model pricing by “Cost per unit of intelligence” rather than “Cost of the very latest thing turned up to 11”, we see that prices are indeed falling.
The trajectory for end-users isn’t perfect. As I’ve said before, we’re in a painful phase now. Frontier labs have been subsidising their models so users get hooked, and they’re raising their prices now, passing on more of the actual cost. But we are getting another glimpse of the future here. Open Weight models (like the Kimi K2.5 that Composer is built upon) are constantly closing the gap with the frontier. Especially if we measure real utility, not just benchmarks.
I expect this trend to continue.