Instruction Following Benchmark (IFBench)
Flint leads the IFBench leaderboard on BenchLM's October 2026 update with 89.2%, ahead of MAI-Thinking-1 (85%) and Qwen3.8 Max (82.8%), across 57 models.
Data verified
IFBench evaluates precise instruction-following generalization on 58 challenging, verifiable out-of-domain constraints. Unlike IFEval which tests familiar constraint types, IFBench specifically measures how well models follow novel instructions they haven't been optimized for, exposing overfitting to common instruction patterns.
Top models on IFBench — October 29, 2026
As of October 29, 2026, Flint leads the IFBench leaderboard with 89.2% , followed by MAI-Thinking-1 (85%) and Qwen3.8 Max (82.8%).
1Open weights
Flint
Recursive
89.2%
Context 256K
2Closed
MAI-Thinking-1
Microsoft
85%
Overall 53.02Context 256K
3Open weights
Qwen3.8 Max
Alibaba
82.8%
Overall 64.86Context 1M
57 modelsInstruction Following70% of category scoreCurrentUpdated October 29, 2026
Leaderboard (57 models)
57 of 57 rows
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | FlintRecursive | 89.2% | Not reported | Open |
| 2 | MAI-Thinking-1Microsoft | 85% | Not reported | Closed |
| 3 | Qwen3.8 MaxAlibaba | 82.8% | Not reported | Open |
| 4 | Inkling-SmallThinking Machines Lab | 82.2% | Not reported | Open |
| 5 | Nemotron 3 UltraNVIDIA | 81.7% | Not reported | Open |
| 6 | Sage-A1.5InternScience | 81.6% | Not reported | Open |
| 7 | Qwen3.8-Omni-FlashAlibaba | 81.5% | Not reported | Closed |
| 8 | Grok 4.3xAI | 81.3% | Not reported | Closed |
| 9 | Qwen3.8-Flash-NextAlibaba | 81.3% | Not reported | Open |
| 10 | Beam 1.0 501B-A23B FP8Reflection AI | 80.7% | Not reported | Open |
| 11 | dots3-note PreviewDots Studio | 80.4% | Not reported | Open |
| 12 | Solar Open 2Upstage | 80% | Not reported | Open |
| 13 | InklingThinking Machines Lab | 79.8% | Not reported | Open |
| 14 | Beam 1.0 501B-A23BReflection AI | 79.6% | Not reported | Open |
| 15 | Qwen3.8-27BAlibaba | 79.5% | Not reported | Open |
| 16 | Granite 4.2 8BIBM | 79.3% | Not reported | Open |
| 17 | Beam 1.0 501B-A23B NVFP4Reflection AI | 79.2% | Not reported | Open |
| 18 | Qwen3.7 MaxAlibaba | 79.1% | Not reported | Closed |
| 19 | Qwen3.7 PlusAlibaba | 79.1% | Not reported | Closed |
| 20 | Granite 4.2 30BIBM | 77.2% | Not reported | Open |
| 21 | Muse Glimmer 30BMeta | 77% | Not reported | Open |
| 22 | Mercury 2.5Inception | 77% | Not reported | Closed |
| 23 | Mercury 3Inception | 76.8% | Not reported | Closed |
| 24 | Qwen3.5 397BAlibaba | 76.5% | Not reported | Open |
| 25 | Gemini 3.5 FlashGoogle | 76.3% | Not reported | Closed |
| 26 | A.X K2SK Telecom | 75.9% | Not reported | Open |
| 27 | Qwen3.6 PlusAlibaba | 75.8% | Not reported | Closed |
| 28 | Spark-X2.6-4BSparkLLM | 75.8% | Not reported | Open |
| 29 | LFM2.5-24B-A2BLiquidAI | 74.7% | Not reported | Open |
| 30 | Ling 3.0 FlashInclusionAI | 74.5% | Not reported | Open |
| 31 | Granite 4.2 3BIBM | 74.3% | Not reported | Open |
| 32 | Nemotron 3 Nano Omni 30B A3BNVIDIA | 74.2% | Not reported | Open |
| 33 | Ternary Bonsai 2 27BPrism ML | 74% | Not reported | Open |
| 34 | Ling 3.0 Flash FP8InclusionAI | 73.4% | Not reported | Open |
| 35 | LFM3-1.7BLiquidAI | 73.0% | Not reported | Open |
| 36 | Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA | 72.9% | Not reported | Open |
| 37 | K-EXAONE 2.0LG AI Research | 72.6% | Not reported | Open |
| 38 | Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIA | 71.5% | Not reported | Open |
| 39 | Mercury 2Inception | 71.3% | Not reported | Closed |
| 40 | Nemotron 3.5 Super VL 120B A12BNVIDIA | 70.9% | Not reported | Open |
| 41 | Agents-A1-4BInternScience | 69.1% | Not reported | Open |
| 42 | MiniCPM5-2BOpenBMB | 66.3% | Not reported | Open |
| 43 | Hy3 PreviewTencent | 63.1% | Not reported | Open |
| 44 | LFM2.5-2.6BLiquidAI | 59.2% | Not reported | Open |
| 45 | Claude Opus 4.5Anthropic | 58% | Not reported | Closed |
| 46 | Ling 2.6 FlashInclusionAI | 57% | Not reported | Open |
| 47 | LFM2.5-8B-A1BLiquidAI | 56.5% | Not reported | Open |
| 48 | OrcaCyber-Nano-1.0OrcaRouter | 56.3% | Not reported | Open |
| 49 | Solar Pro 3Upstage | 55.8% | Not reported | Closed |
| 50 | ZAYA1-8BZyphra | 52.6% | Not reported | Open |
| 51 | Trinity-Large-ThinkingArcee AI | 52.3% | Not reported | Open |
| 52 | MiniCPM5-1BOpenBMB | 46.7% | Not reported | Open |
| 53 | LFM2.5-230MLiquidAI | 38.4% | Not reported | Open |
| 54 | Kanana-2 1.3B InstructKakao | 34.7% | Not reported | Open |
| 55 | Kanana-2 3B InstructKakao | 33.3% | Not reported | Open |
| 56 | LFM2.5-VL-3BLiquidAI | 25.8% | Not reported | Open |
| 57 | LLaDA2.2-miniInclusionAI | 24.9% | Not reported | Open |
AA-IFBench
BenchLM stores the Artificial Analysis IFBench result separately from the core IFBench lane because Artificial Analysis controls the evaluation configuration.
165 of 165 rows
| Source rank | Source model or configuration | Score | Source setup | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | MiniMax M3MiniMax | 82.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 2 | Nemotron 3 UltraNVIDIA | 81.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 3 | Grok 4.3xAI | 81.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 4 | Qwen3.7 MaxAlibaba | 80.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 5 | MiMo-V2.5-ProXiaomi | 79.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 6 | Qwen3.5 397BAlibaba | 78.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 7 | Qwen3.5 397B (Reasoning)Alibaba | 78.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 8 | Qwen3.7 PlusAlibaba | 78.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 9 | GPT-5.2-CodexOpenAI | 77.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 10 | Gemini 3.1 ProGoogle | 77.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 11 | Qwen 3.6 Max (preview)Alibaba | 76.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 12 | DeepSeek V4 Pro 0813DeepSeek | 76.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 13 | Gemini 3.5 FlashGoogle | 76.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 14 | GLM-5.1Z.AI | 76.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 15 | Kimi K2.6Moonshot AI | 76.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 16 | GPT-5.5OpenAI | 75.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 17 | Muse SparkMeta | 75.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 18 | GPT-5.4 nanoOpenAI | 75.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 19 | Qwen3.5-122B-A10BAlibaba | 75.7% | Artificial AnalysisOriginal result | Not reported | Open |
| 20 | MiniMax M2.7MiniMax | 75.7% | Artificial AnalysisOriginal result | Not reported | Open |
| 21 | Qwen3.5-27BAlibaba | 75.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 22 | Gemma 4 31BGoogle | 75.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 23 | GPT-5.3 CodexOpenAI | 75.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 24 | GPT-5.2OpenAI | 75.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 25 | GPT-5.3-Codex-SparkOpenAI | 75.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 26 | GPT-5 miniOpenAI | 75.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 27 | Qwen3.6 PlusAlibaba | 75.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 28 | GPT-5.4OpenAI | 73.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 29 | Command A+Cohere | 73.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 30 | Gemma 4 12BGoogle | 73.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 31 | GLM-5.2Z.AI | 73.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 32 | GPT-5.4 miniOpenAI | 73.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 33 | GLM-5-TurboZ.AI | 73.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 34 | GPT-5 (high)OpenAI | 73.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 35 | GPT-5.1OpenAI | 72.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 36 | GPT-5.6 SolOpenAI | 72.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 37 | Qwen3.5-35B-A3BAlibaba | 72.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 38 | Gemma 4 26B A4BGoogle | 72.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 39 | GLM-5Z.AI | 72.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 40 | MiniMax M2.5MiniMax | 71.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 41 | o3OpenAI | 71.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 42 | DeepSeek V4 Pro (High)DeepSeek | 71.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 43 | GPT-5.6 TerraOpenAI | 71.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 44 | Solar Pro 3Upstage | 71.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 45 | GPT-5 (medium)OpenAI | 70.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 46 | Gemini 3 ProGoogle | 70.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 47 | o1OpenAI | 70.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 48 | Kimi K2.5 (Reasoning)Moonshot AI | 70.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 49 | Kimi K2.5Moonshot AI | 70.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 50 | GPT-5.1-CodexOpenAI | 70.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 51 | GPT-5.1-Codex-MaxOpenAI | 70.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 52 | Mercury 2Inception | 69.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 53 | GPT-OSS 120BOpenAI | 69.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 54 | MiMo-V2-ProXiaomi | 68.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 55 | Mistral Medium 3.5 128BMistral | 68.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 56 | o4-miniOpenAI | 68.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 57 | GLM-4.7Z.AI | 67.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 58 | Qwen3.6-27BAlibaba | 67.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 59 | GPT-5 nanoOpenAI | 67.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 60 | Step 3.7 FlashStepFun | 67.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 61 | Step 3.5 FlashStepFun | 66.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 62 | GPT-OSS 20BOpenAI | 65.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 63 | K-ExaoneLG AI Research | 64.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 64 | Qwen3.6-35B-A3BAlibaba | 64.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 65 | Claude Fable 5Anthropic | 63.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 66 | Nemotron 3 Nano Omni 30B A3BNVIDIA | 63.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 67 | Kimi K2.7 CodeMoonshot AI | 63.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 68 | Claude Opus 4.8Anthropic | 62.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 69 | GLM-5V-TurboZ.AI | 61.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 70 | Nemotron 3.5 Bolt E4BNVIDIA | 61.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 71 | GLM-4.7-FlashZ.AI | 60.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 72 | Claude Opus 4.7 (Adaptive)Anthropic | 58.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 73 | Claude Opus 4.5 ThinkingAnthropic | 58.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 74 | North Mini CodeCohere | 57.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 75 | Ling 2.6 FlashInclusionAI | 57.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 76 | Trinity-Large-ThinkingArcee AI | 56.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 77 | Trinity-Large-PreviewArcee AI | 56.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 78 | LFM2.5-8B-A1BLiquidAI | 55.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 79 | Claude 4.1 Opus ThinkingAnthropic | 55.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 80 | Gemini 3 FlashGoogle | 55.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 81 | Claude 4 Opus ThinkingAnthropic | 53.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 82 | Grok 4xAI | 53.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 83 | MiMo-V2-OmniXiaomi | 53.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 84 | Claude Opus 4.6 (Adaptive)Anthropic | 53.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 85 | Grok 4.1 Fast (Reasoning)xAI | 52.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 86 | Qwen3.5-4BAlibaba | 52.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 87 | Grok 4 Fast (Reasoning)xAI | 50.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 88 | DeepSeek V3.2DeepSeek | 49.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 89 | Gemini 2.5 ProGoogle | 48.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 90 | Claude 3.7 Sonnet ThinkingAnthropic | 48.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 91 | Mistral Small 4Mistral | 48.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 92 | Llama 3.3 70BMeta | 47.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 93 | Ultravox v0.6 Llama 3.3 70BFixie AI | 47.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 94 | LFM2-24B-A2BLiquidAI | 45.9% | Artificial AnalysisOriginal result | Not reported | Closed |
| 95 | Claude 4 SonnetAnthropic | 45.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 96 | Claude Opus 4.6Anthropic | 44.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 97 | Gemma 4 E4BGoogle | 44.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 98 | Qwen3 MaxAlibaba | 44.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 99 | Claude 3.7 SonnetAnthropic | 44.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 100 | LFM2.5-1.2B-InstructLiquidAI | 43.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 101 | Claude Opus 4.7Anthropic | 43.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 102 | Qwen3-Omni-30B-A3B-ThinkingAlibaba | 43.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 103 | Claude 4 OpusAnthropic | 43.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 104 | Claude Opus 4.5Anthropic | 43.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 105 | GPT-4.1OpenAI | 43.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 106 | Llama 4 MaverickMeta | 43.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 107 | Claude 3.5 HaikuAnthropic | 42.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 108 | MiniMax M1 80kMiniMax | 41.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 109 | LFM2.5-1.2B-ThinkingLiquidAI | 41.8% | Artificial AnalysisOriginal result | Not reported | Closed |
| 110 | DeepSeek V3.1 (Reasoning)DeepSeek | 41.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 111 | Kimi K2Moonshot AI | 41.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 112 | Grok Code Fast 1xAI | 41.4% | Artificial AnalysisOriginal result | Not reported | Closed |
| 113 | Claude Sonnet 4.6Anthropic | 41.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 114 | DeepSeek V3 0324DeepSeek | 41.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 115 | Gemini 2.0 FlashGoogle | 40.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 116 | MiMo-V2-FlashXiaomi | 39.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 117 | DeepSeek-R1DeepSeek | 39.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 118 | Llama 4 ScoutMeta | 39.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 119 | Mistral Medium 3Mistral | 39.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 120 | Gemini 2.5 FlashGoogle | 39.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 121 | Llama 3.1 405BMeta | 39.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 122 | QwQ-32BAlibaba | 38.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 123 | GPT-4.1 miniOpenAI | 38.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 124 | Nemotron Ultra 253BNVIDIA | 38.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 125 | Nova ProAmazon | 38.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 126 | Gemma 4 E2BGoogle | 38.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 127 | DeepSeek V3.1DeepSeek | 37.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 128 | GLM-4.5-AirZ.AI | 37.6% | Artificial AnalysisOriginal result | Not reported | Closed |
| 129 | Nemotron 3 Nano 30BNVIDIA | 37.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 130 | GLM-4.6Z.AI | 36.7% | Artificial AnalysisOriginal result | Not reported | Open |
| 131 | Grok 4.1 FastxAI | 36.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 132 | Mistral Large 3Mistral | 36.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 133 | Claude 3 HaikuAnthropic | 36.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 134 | DeepSeek V3DeepSeek | 34.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 135 | Devstral SmallMistral | 34.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 136 | Sarvam 105BSarvam | 34.4% | Artificial AnalysisOriginal result | Not reported | Open |
| 137 | GPT-4oOpenAI | 34.3% | Artificial AnalysisOriginal result | Not reported | Closed |
| 138 | Solar Pro 2Upstage | 33.7% | Artificial AnalysisOriginal result | Not reported | Closed |
| 139 | Exaone 4.0 32BLG AI Research | 33.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 140 | LFM2.5-VL-1.6B-ExtractLiquidAI | 33.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 141 | Mistral Small 4 (Reasoning)Mistral | 32.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 142 | GPT-4.1 nanoOpenAI | 32.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 143 | Ministral 3 14B (Reasoning)Mistral | 32.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 144 | Ministral 3 14BMistral | 32.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 145 | Gemma 3 27BGoogle | 31.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 146 | Gemini 2.5 Flash-LiteGoogle | 31.5% | Artificial AnalysisOriginal result | Not reported | Closed |
| 147 | Mistral Large 2Mistral | 31.2% | Artificial AnalysisOriginal result | Not reported | Closed |
| 148 | Qwen3-Omni-30B-A3B-InstructAlibaba | 31.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 149 | GPT-4o miniOpenAI | 31.0% | Artificial AnalysisOriginal result | Not reported | Closed |
| 150 | Mistral Small 3.1Mistral | 29.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 151 | Ministral 3 8B (Reasoning)Mistral | 29.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 152 | Ministral 3 8BMistral | 29.1% | Artificial AnalysisOriginal result | Not reported | Open |
| 153 | Llama 3.1 70BMeta | 29.0% | Artificial AnalysisOriginal result | Not reported | Open |
| 154 | Ministral 3 3B (Reasoning)Mistral | 26.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 155 | Ministral 3 3BMistral | 26.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 156 | Sarvam 30BSarvam | 26.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 157 | Granite-4.0-H-1BIBM | 26.2% | Artificial AnalysisOriginal result | Not reported | Open |
| 158 | Exaone 4.0 1.2BLG AI Research | 25.3% | Artificial AnalysisOriginal result | Not reported | Open |
| 159 | Magistral MediumMistral | 25.1% | Artificial AnalysisOriginal result | Not reported | Closed |
| 160 | Magistral SmallMistral | 24.8% | Artificial AnalysisOriginal result | Not reported | Open |
| 161 | Phi-4Microsoft | 23.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 162 | Phi-4-reasoningMicrosoft | 23.5% | Artificial AnalysisOriginal result | Not reported | Open |
| 163 | DeepSeek R1 Distill Qwen 32BDeepSeek | 22.9% | Artificial AnalysisOriginal result | Not reported | Open |
| 164 | Granite-4.0-H-350MIBM | 17.6% | Artificial AnalysisOriginal result | Not reported | Open |
| 165 | Granite-4.0-350MIBM | 15.9% | Artificial AnalysisOriginal result | Not reported | Open |
According to BenchLM.ai, Flint leads the IFBench benchmark with a score of 89.2%, followed by MAI-Thinking-1 (85%) and Qwen3.8 Max (82.8%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
57 models have been evaluated on IFBench. The benchmark falls in the Instruction Following category. The Instruction Following leaderboard ranks models by a weighted category score, and IFBench contributes 70% of it. BenchAlign v5.8 also uses it in the overall ranking.
About IFBench
Year
2025
Tasks
58
Freshness and provenance
Version
IFBench 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
Current
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does IFBench measure?
IFBench evaluates precise instruction-following generalization on 58 challenging, verifiable out-of-domain constraints. Unlike IFEval which tests familiar constraint types, IFBench specifically measures how well models follow novel instructions they haven't been optimized for, exposing overfitting to common instruction patterns.
Which model scores highest on IFBench?
Flint by Recursive currently leads with a score of 89.2% on IFBench.
How many models are evaluated on IFBench?
57 AI models have published results on IFBench in the BenchLM catalog.
Compare top models on IFBench
Last updated: October 29, 2026 · BenchLM version IFBench 2025
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.