FA Dev Tools

AI Models

Compare 20 model versions — Claude Opus, Sonnet and Fable, Grok, DeepSeek and Cursor Composer, with GPT and Gemini for reference — on API price, what Cursor charges, benchmarks from the makers and from independent leaderboards, and articles that put them head to head.

Quick compare

Side by side

Pick a quick compare above, or tick models in the list below.

Articles and comparisons

42 reviews, head-to-heads and benchmark write-ups, newest first.

Families — none picked means all
Search

All models

12 of 20 versions. Tick up to 8 to compare.

Sort by
ModelReleasedAPI in / outCursor in / outContextAA IndexArena Elo
2026-09-22$2 / $10—1.05M48—
2026-09-21$2 / $6$2 / $6500K46—
2026-09-10$0.30 / $1.20—1M39—
2026-09-04$10 / $50—1.05M531480
2026-09-02$0.75 / $3.75$0.75 / $3.501.05M411493
2026-08-13$1.32 / $3.96—1M361463
2026-08-13$0.75 / $3.75$0.75 / $3.501.05M—1490
2026-08-12$2 / $6$2 / $6500K441456
2026-07-09$4 / $20$4 / $201.05M—1483
2026-05-29$1 / $2—256K——
2026-05-18Cursor only$0.50 / $2.50200K——
2026-02-19$2 / $12$2 / $121.05M301487

Reading the numbers

Checked September 2026. Prices are official US list prices per million tokens; a reported tag means the maker no longer lists it (retired versions) and it comes from trackers. OpenRouter now is fetched live and can be lower than list price when a cheaper host serves the model.

In Cursor is what Cursor bills against your plan. Composer and Grok sit in Cursor's own pool, which every plan includes much more of; everything else is charged at the API price. OpenAI is cutting Cursor off from its models on 12 November 2026.

Benchmarks marked vendor were run by the model's maker, on its own harness and effort setting; independent runs (Artificial Analysis, Vals, arena.ai, ARC Prize, tbench.ai) often land lower. Each number links its source. Versions are separate rows — Terminal-Bench 2.0, 2.1 and 4.0 are different task sets and can't be compared. Most labs stopped publishing SWE-bench Verified, GPQA and AIME in 2026 because they are saturated, so newer models have gaps there rather than low scores.

Instructions

Compare AI models version by version — Claude Opus and Sonnet, Grok, DeepSeek and Cursor Composer, plus GPT and Gemini — on price, benchmarks and what reviewers wrote.

  1. Click a Quick compare such as Composer 2.5 vs Opus 5, or tick models in All models.
  2. Read Side by side: API price, the price inside Cursor, a live OpenRouter price, then benchmarks grouped into overall, coding and reasoning. The best value in each row is green.
  3. Articles about these models lists head-to-head reviews first, then ones comparing the same families at other versions.
  4. Narrow the list with the Families chips or the search box, and tick Include legacy and retired versions to see older releases.

Tips

  • Blended (3:1) weights input three times output, a rough stand-in for a coding session. Composer has no public API, so its blend uses Cursor's price.
  • Hover a benchmark score to see who measured it and at what effort; the row name links the leaderboard.
  • Your picks and filters are kept in local storage on this device only. The live price comes from OpenRouter via this site.
  • Hit Example and click the highlighted Composer 2.5 vs Opus 5 quick compare to see a full comparison with its articles.