Cursor

• No official SWE-bench score published. Independent tests with a Claude Sonnet backend measured 55–62%. Cursor is faster on small, focused tasks and wins on generation speed for inline edits. • $20 Pro is a credit pool; heavy premium-model users report $10–20/day in overages. But on simple utility tasks Cursor delivers about 42 accuracy points per dollar vs Claude Code's 31 — the economics flip on high-frequency simple work. • Model-dependent: ~128K with GPT-class models, up to 200K with Claude backends; independent audits report ~70K–120K usable after truncation and prompt overhead, with RAG chunking covering the rest. • OpenAI announced on 2026-08-28 that it notified SpaceX (which acquired Cursor's parent Anysphere for ~$60B) of a proposed 2026-11-12 termination — the longest notice the contract allows. No new OpenAI models ship to Cursor before then; existing ones run to the cutoff. BYOK with your own OpenAI key is reported to keep working, and OpenAI models are only about 5% of Cursor traffic. Anthropic is reported to be increasing compute support, without a confirmed public plan. • VS Code fork with the strongest inline Tab completion in 2026, colour-coded visual diffs, and the entire VS Code extension ecosystem — productive within minutes for anyone who already uses VS Code, with a free tier to start. • OpenAI notified Cursor on 2026-08-28 of a planned 2026-11-12 end to direct model supply (the longest notice the contract allows), with no new models supplied beforehand. The reported trigger is SpaceX's acquisition of Anysphere (~$60B all-stock, early Aug 2026). Key distinction: this cuts the official direct contract — BYOK is reportedly still usable, and OpenAI models are ≈5% of Cursor traffic. Anthropic was reported to be adding capacity, with no publicly confirmed arrangement. • Now sold as part of The X Subscription across five tiers: Free $0, Plus $8/mo, 高级 (Premium) $30/mo, Super $100/mo, Ultra $200/mo; Teams $30 / $100 / $200 per seat per month; Enterprise custom. Annual: Plus $80/yr, 高级 $300/yr, Super $1,000/yr, Ultra $2,000/yr. The included usage is a weekly pool shared with Grok. Critically, third-party models (Claude, GPT, Gemini) are not in the included pool — from Premium up they are billed at API list price against prepaid credits. Plus and Free have no third-party models at all. • Ships its own editor built as a VS Code fork, plus a CLI and a Slack integration. Deeply integrated, but you work inside Cursor's app — there is no JetBrains build. • Third-party models now metered at API pricing instead of included; no JetBrains support; OpenAI direct supply faces a planned 2026-11-12 cutoff. • $20/mo Pro; 500 advanced requests then extra; more generous free Hobby. • 30% faster task completion (62.9s); 81% multi-file accuracy; 1M token context. • Composer coordinates 10-50 files; autonomy slider; understands whole codebase. • VS Code branch (zero migration); AI-native; JetBrains support still beta. • OpenAI announced on 2026-08-28 that it will stop supplying models directly to Cursor, with a proposed end date of 2026-11-12 — OpenAI called this the longest notice period the contract allows. What is being cut is Cursor's official direct-supply contract (Cursor buying models in bulk and bundling them into its subscription). Bring-your-own-key with a personal OpenAI API key is reported to remain usable, and OpenAI's own Cursor extension keeps working. However, future OpenAI models — including the next flagship (GPT-6 Astra) — will not be provided to Cursor.