Cursor ships Composer 2.5 — in-house long-horizon model matches Opus 4.7 and GPT-5.5 on benchmarks at $0.50/$2.50 per million tokens
Cursor shipped Composer 2.5 in May 2026 — an in-house long-horizon coding model that matches Opus 4.7 and GPT-5.5 on benchmarks at $0.50 input / $2.50 output per million tokens. The pricing is roughly an order of magnitude below the comparable frontier-lab API rates, and the strategic implication is that Cursor's economics now hold together without requiring third-party model spend.
The in-house model economics matter because the prior Cursor business model depended on subsidized third-party model access. Cursor pays Anthropic and OpenAI for the model calls its users make; the $20/month Cursor Pro tier covers a usage envelope, beyond which the per-token cost becomes the constraint. With Composer 2.5 matching Opus 4.7 and GPT-5.5 at one-tenth the cost, Cursor can serve high-volume professional users on its own model without margin erosion — and route through the frontier-lab models only for edge cases where Composer 2.5 underperforms. That's a fundamentally different cost structure than the all-third-party Cursor of 2025.
Composer 2.5 also adds two productivity surfaces: Build in Parallel (which dispatches multiple competing implementations of a task and merges the best) and PR review (Cursor-native code review that operates against the user's branch). Both are explicit plays at workflows that GitHub Copilot and Claude Code currently serve. With its own model and competitive feature parity, Cursor's $20/month positioning becomes harder to displace — the price/feature trade-off advantages of an outside competitor have to overcome an IDE-plus-model integration that the user already pays for and uses every day.
Nipralo Technologies — Best AI Coding Tools 2026 Claude Code Cursor Copilot → · Lushbinary — AI Coding Agents 2026 Cursor Claude Copilot Comparison → · SitePoint — Claude Code vs Cursor vs Copilot 2026 Comparison →