Benchmarks / Long Context & Instruction
IFBench
Instruction-following benchmark (verifiable constraints).
maintainer unknown · 207 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| IFBench current ifbench | — |
Current leaderboard
IFBench · Accuracy · top 25 of 207 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | Grok 4.3medium effort xAI | 83.3% |
| 2 | Minimax M3 MiniMax | 82.9% |
| 3 | Nemotron 3 Ultra 550B A55bthinking NVIDIA | 81.4% |
| 4 | Grok 4.3high effort xAI | 81.3% |
| 5 | Grok 4.3low effort xAI | 81.0% |
| 6 | Qwen3 7 Max Alibaba (Qwen) | 80.5% |
| 7 | Mimo V2 5 Pro xiaomi | 79.9% |
| 8 | Deepseek V4 Flashmax effort DeepSeek | 79.2% |
| 9 | Qwen3 5 397B A17bthinking Alibaba | 78.8% |
| 10 | Qwen3 7 Plus Alibaba | 78.0% |
| 11 | Gemini 3 Flash Previewthinking Google DeepMind | 78.0% |
| 12 | GPT 5.2 Codexxhigh effort OpenAI | 77.6% |
| 13 | Gemini 3.1 Flash Lite Preview Google | 77.2% |
| 14 | Gemini 3.1 Pro Preview Google DeepMind | 77.1% |
| 15 | Deepseek V4 Promax effort DeepSeek | 76.5% |
| 16 | Gemini 3.5 Flashhigh effort Google DeepMind | 76.3% |
| 17 | GLM 5.1thinking Z.ai | 76.3% |
| 18 | Kimi K2 6 Moonshot AI | 76.0% |
| 19 | Muse Spark Meta AI | 75.9% |
| 20 | GPT 5.4 Nano (2026-03-17)xhigh effort OpenAI | 75.9% |
| 21 | GPT 5.5 (2026-04-22)xhigh effort OpenAI | 75.8% |
| 22 | Qwen3 5 122B A10bthinking Alibaba | 75.7% |
| 23 | Minimax M2 7 MiniMax | 75.7% |
| 24 | Qwen3 5 27Bthinking Alibaba | 75.6% |
| 25 | Gemma 4 31Bthinking Google DeepMind | 75.6% |
One row per evaluated system — reasoning-effort variants rank separately.