Cursor Launches Composer 2 at 86% Lower Cost, Challenging Claude and GPT-5.4
The Development
Anysphere, the $29.3 billion San Francisco startup behind Cursor, launched Composer 2 on March 19, 2026—its most capable in-house coding model to date. Composer 2 Standard is priced at $0.50 per million input tokens and $2.50 per million output tokens, an 86% reduction from Composer 1.5's $3.50/$17.50 pricing. The faster Composer 2 Fast tier, at $1.50/$7.50, is now the default experience. On Cursor's own CursorBench, Composer 2 scores 61.3, up from Composer 1.5's 44.2. On SWE-bench Multilingual it scores 73.7 versus 65.9 prior. It beats Claude Opus 4.6 across published benchmarks but trails GPT-5.4 on Terminal-Bench 2.0 (61.7 vs. 75.1). The model is not available as a standalone API—it is exclusive to Cursor's agentic environment, integrated with file editing, semantic code search, shell commands, and browser control.
Our Take
The 86% price cut is a defensive move, not a victory lap. Cursor's core business model—charging teams $40 per user per month to sit between developers and foundation models—is under direct assault from Anthropic's Claude Code and OpenAI's Codex, both terminal-native, first-party products that cut out the intermediary entirely. By releasing Composer 2 as a Cursor-exclusive model, Anysphere is betting that tight platform integration and lower costs outweigh the gravitational pull of working directly with model makers. That's a viable argument for enterprise buyers who value governance, pooled usage, and SSO. It's a weaker argument for power developers, who are already migrating toward Claude Code's leaner, fully agentic workflow. Cursor's window to lock in enterprise accounts before first-party tools mature is Q2 2026. Composer 2's economics buy time—but not a structural moat.
The Exploit
Action Item
If your engineering organization is currently evaluating AI coding platforms for an enterprise contract, use Composer 2's launch pricing as direct leverage to negotiate Cursor's Teams or Enterprise tier before first-party alternatives from Anthropic and OpenAI formalize their own team-tier pricing in Q2 2026.