List your company
ResearchFormerly WizWand

Instruction Following on IFBench (Score %)

82IFBench ScoreDS-v4-Flash
13.786431.495749.20566.9143Nov 28, 2025Dec 30, 2025Feb 1, 2026Mar 6, 2026Apr 8, 2026May 11, 2026Jun 13, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DS-v4-FlashPrompting protocol=prompt loose, Parameter Count=284B-A13B2026.0682
Nemotron 3 UltraPrompting protocol=prompt loose, Parameter Count=550B-A55B2026.0681.7
DS-v4-ProPrompting protocol=prompt loose, Parameter Count=1.6T-A49B2026.0679.1
Qwen-3.5Prompting protocol=prompt loose, Parameter Count=397B-17B2026.0678.2
GLM-5.1Prompting protocol=prompt loose, Parameter Count=744B-A40B2026.0676.6
DeepSeek-V4-ProThinking mode=true2026.0576
MiniMax-2.7Prompting protocol=prompt loose, Parameter Count=230B-A10B2026.0674.6
GLM5.1Thinking mode=true2026.0574
Kimi-K2.6Prompting protocol=prompt loose, Parameter Count=1T-A32B2026.0673.7
Gemini 3.1 ProThinking mode=true2026.0571.33
Kimi K2.6Thinking mode=true2026.0568.67
GPT-5.5Thinking mode=true2026.0567.33
Qwen3.5-397B-A17BThinking mode=true2026.0565.67
TeacherDistillation Setting=Multi-Domain Distillation, Teacher Model=Qwen3-Nemotron-4B2026.0562.93
Qwen3.5-4BActive / Total Parameters=4.0B / 4.0B, Sampling Settings=Recommended sampling settings2026.0559.2
Ling-2.6-1T2026.0657.62
GLM-5Thinking Mode=non-thinking2026.0657.14
Qwen3-4B-Thinking-2507Active / Total Parameters=4.0B / 4.0B, Sampling Settings=Recommended sampling settings2026.0552.9
ZAYA1-8BActive / Total Parameters=0.7B / 8.0B, Sampling Settings=T=1.0, top-p=0.95, top-k disabled2026.0552.6
Hy-MT2-30B-A3BThinking mode=false2026.0550.67
GLM5.1Thinking mode=false2026.0550
DeepSeek-V3.2Thinking Mode=nothink2026.0650
GPT-5.4Reasoning Mode=non-reasoning2026.0649.73
Qwen3.5-397B-A17BThinking mode=false2026.0548.33
Gemma4-26B-A4BThinking mode=false2026.0545.6
GPT-5.5Thinking mode=false2026.0543.33
STEP3-VL-10BNumber of Parameters=10B2026.0143.28
Gemma-4-E4B-itActive / Total Parameters=4.0B / 8.0B, Sampling Settings=Recommended sampling settings2026.0542.7
DeepSeek-V4-ProThinking mode=false2026.0542.67
Kimi-K2.5Mode=Instant2026.0642.53
TrOPDDistillation Setting=Multi-Domain Distillation, Student Model=Qwen3-SFT-1.7B, Teacher Model=Qwen3-Nemotron-4B2026.0542.18
HeRLBackbone=Qwen2.5-7B-Instruct, Training Method=HeRL2026.0339.7
HeRLBackbone=Qwen3-4B-Instruct-2507, Training Method=HeRL2026.0339.7
Entropy OPDDistillation Setting=Multi-Domain Distillation, Student Model=Qwen3-SFT-1.7B, Teacher Model=Qwen3-Nemotron-4B2026.0538.78
Kimi K2.6Thinking mode=false2026.0538
OPDDistillation Setting=Multi-Domain Distillation, Student Model=Qwen3-SFT-1.7B, Teacher Model=Qwen3-Nemotron-4B2026.0537.07
RLVRBackbone=Qwen3-4B-Instruct-2507, Training Method=RLVR2026.0336.9
Qwen3-VL ThinkingNumber of Parameters=8B2026.0136.65
EOPDDistillation Setting=Multi-Domain Distillation, Student Model=Qwen3-SFT-1.7B, Teacher Model=Qwen3-Nemotron-4B2026.0536.39
REOPOLDDistillation Setting=Multi-Domain Distillation, Student Model=Qwen3-SFT-1.7B, Teacher Model=Qwen3-Nemotron-4B2026.0536.05
Qwen3.6-35B-A3BThinking mode=false2026.0536
Hy-MT2-1.8BThinking mode=false2026.0535.33
Hy-MT2-7BThinking mode=false2026.0535.33
Gemma4-E4BThinking mode=false2026.0532
RLVRBackbone=Qwen2.5-7B-Instruct, Training Method=RLVR2026.0331.6
SFTBackbone=Qwen3-4B-Instruct-2507, Training Method=SFT2026.0331.3
HeRLBackbone=Llama-3.2-3B-Instruct, Training Method=HeRL2026.0330.6
Qwen3-4B-Instruct-2507Backbone=Qwen3-4B-Instruct-2507, Training Method=Initial2026.0329.9
InternVL 3.5Number of Parameters=8B2026.0128.14
SFTBackbone=Qwen2.5-7B-Instruct, Training Method=SFT2026.0327.9
DPOBackbone=Qwen3-4B-Instruct-2507, Training Method=DPO2026.0327.9
GLM-4.6V FlashNumber of Parameters=9B2026.0127.47
RLVRBackbone=Llama-3.2-3B-Instruct, Training Method=RLVR2026.0326.6
Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct, Training Method=Initial2026.0326.2
Qwen3-SFT-1.7BDistillation Setting=Baseline (None), Student Model=Qwen3-SFT-1.7B2026.0526.19
Gemma4-E2BThinking mode=false2026.0526
DPOBackbone=Qwen2.5-7B-Instruct, Training Method=DPO2026.0325.9
SFTBackbone=Llama-3.2-3B-Instruct, Training Method=SFT2026.0324.8
Llama-3.2-3B-InstructBackbone=Llama-3.2-3B-Instruct, Training Method=Initial2026.0323.8
MiMo-VL RL-2508Number of Parameters=7B2026.0123.47
DPOBackbone=Llama-3.2-3B-Instruct, Training Method=DPO2026.0322.1
Qwen3-1.7B# Total Params=1.7B, # Trained Tokens=36T2025.1121.27
LFM2-1.2B# Total Params=1.2B, # Trained Tokens=11T2025.1120.7
LFM2-700M# Total Params=0.70B, # Trained Tokens=11T2025.1120.56
Qwen3-0.6B# Total Params=0.6B, # Trained Tokens=36T2025.1119.75
Gemma-3-1B# Total Params=1B, # Trained Tokens=2T2025.1117.72
Llama-3.2-1B# Total Params=1.2B, # Trained Tokens=9T2025.1116.86
LFM2-350M# Total Params=0.35B, # Trained Tokens=11T2025.1116.41