CAP
Instruction Following
League table — best at instruction following · 30
mean percentile across this capability's benchmarks| # | Model | League | Coverage | $/1M | tok/s |
|---|---|---|---|---|---|
| 1 | Grok 4.20 0309 | 100.0 | 1/1 | $3 | — |
| 2 | MiniMax-M3 | 99.7 | 1/1 | $0.525 | 87 |
| 3 | Nemotron 3 Ultra 550B A55B | 99.4 | 1/1 | $1.18 | 180 |
| 4 | Grok 4.3 | 99.1 | 1/1 | $1.56 | — |
| 5 | Grok 4.20 0309 v2 | 98.8 | 1/1 | $3 | — |
| 6 | Qwen3.7 Max | 98.5 | 1/1 | $3.75 | 200 |
| 7 | Nemotron Cascade 2 30B A3B | 98.2 | 1/1 | — | — |
| 8 | MiMo-V2.5-Pro | 97.9 | 1/1 | $0.544 | 65 |
| 9 | DeepSeek V4 Flash | 97.6 | 1/1 | $0.175 | 118 |
| 10 | Nova 2.0 Pro Preview | 97.3 | 1/1 | $3.44 | 136 |
| 11 | Qwen3.5 397B A17B | 97.0 | 1/1 | $1.35 | 66 |
| 12 | Gemini 3 Flash Preview | 96.7 | 1/1 | $1.13 | — |
| 13 | Qwen3.7 Plus | 96.4 | 1/1 | $0.7 | 53 |
| 14 | GPT-5.2 Codex | 96.0 | 1/1 | $4.81 | — |
| 15 | Gemini 3.1 Flash-Lite | 95.7 | 1/1 | $0.563 | 299 |
| 16 | Gemini 3.1 Pro Preview | 95.4 | 1/1 | $4.5 | 132 |
| 17 | Qwen3.6 Max Preview | 95.1 | 1/1 | $2.92 | — |
| 18 | DeepSeek V4 Pro | 94.8 | 1/1 | $0.544 | 71 |
| 19 | GLM-5.1 | 94.5 | 1/1 | $2.14 | — |
| 20 | Gemini 3.5 Flash | 94.2 | 1/1 | $3.38 | 250 |
| 21 | Kimi K2.6 | 93.9 | 1/1 | $1.71 | — |
| 22 | GPT-5.4 nano | 93.6 | 1/1 | $0.463 | — |
| 23 | GPT-5.5 | 93.3 | 1/1 | $11.25 | — |
| 24 | Muse Spark | 93.0 | 1/1 | — | — |
| 25 | MiniMax-M2.7 | 92.7 | 1/1 | $0.525 | — |
| 26 | Qwen3.5 122B A10B | 92.4 | 1/1 | $1.1 | 135 |
| 27 | Qwen3.5 27B | 92.1 | 1/1 | $0.825 | — |
| 28 | Gemma 4 31B | 91.8 | 1/1 | — | 35 |
| 29 | GPT-5.2 | 91.5 | 1/1 | $4.81 | — |
| 30 | GPT-5.3 Codex | 91.2 | 1/1 | $4.81 | 126 |
League = a model's mean percentile across the 1 benchmarkin this capability (best score per benchmark; partial coverage shown). Price & speed via Artificial Analysis.
Benchmarks in this capability (1)
IFBench330 scores