Leaderboard
Every model's real record, from actual head-to-head votes. No benchmark, no vendor claim, just what people picked when they saw the answers side by side.
| # | Model | Win Rate | To First Token | Speed |
|---|---|---|---|---|
| 1 | i inclusionAI: Ling 3.0 Tiny (free) | 100%won 2 of 2 | 1673 ms | 449 tok/s |
| 2 | D Dots Studio: Dots3-Note Preview (free) | 100%won 1 of 1 | 2601 ms | 110.1 tok/s |
| 3 | M MiniMax: MiniMax M2.7 (free) | 50%won 1 of 2 | 5466 ms | 46.9 tok/s |
| 4 | P Poolside: Laguna S 2.1 (free) | 0%won 0 of 2 | 7788 ms | 69.4 tok/s |
| 5 | C Cohere: North Mini Code (free) | 0%won 0 of 1 | 3055 ms | 120.9 tok/s |
| 6 | F Free Models Router | 0%won 0 of 1 | 3517 ms | 19.5 tok/s |
| 7 | G Google: Gemma 4 26B A4B (free) | 0%won 0 of 1 | 6369 ms | 56.4 tok/s |
| 8 | O OpenAI: gpt-oss-20b (free) | 0%won 0 of 1 | 4544 ms | 52.8 tok/s |
| 9 | N NVIDIA: Nemotron 3 Nano 30B A3B (free) | 0%won 0 of 0 | 3736 ms | 222.9 tok/s |
| 10 | N NVIDIA: Nemotron 3 Super (free) | 0%won 0 of 0 | 36558 ms | 774.1 tok/s |
| 11 | N NVIDIA: Nemotron 3 Ultra (free) | 0%won 0 of 0 | 9220 ms | 27.7 tok/s |
Speed is wall clock, request to finish, so a model that buffers its whole answer and a model that streams token by token can sit in the same column honestly. Time to first token is measured separately and shown beside it.