Visual Question Answering on ChartQA
90.5AccuracyPenguin-VL
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Penguin-VLModel Size=8B2026.03 | 90.5 | — | — | — | — | — | — | — | |
| QATModel=Llama Nemotron Nano 12B v2 VL2026.01 | 89.8 | — | — | — | — | — | — | — | |
| BaselineModel=Llama Nemotron Nano 12B v2 VL2026.01 | 89.7 | — | — | — | — | — | — | — | |
| PTQModel=Llama Nemotron Nano 12B v2 VL2026.01 | 89.6 | — | — | — | — | — | — | — | |
| Qwen3-VLModel Size=8B2026.03 | 89.6 | — | — | — | — | — | — | — | |
| Qwen3-VL-8B2026.07 | 89.6 | — | — | — | — | — | — | — | |
| Qwen2.5VL 7B# Vis tok.=14002025.04 | 89.5 | — | — | — | — | — | — | — | |
| Kimi-VL-A3B-Instructk=6, Status=Default2025.11 | 89.48 | — | — | — | — | — | — | — | |
| QADModel=Llama Nemotron Nano 12B v2 VL2026.01 | 89.4 | — | — | — | — | — | — | — | |
| MoDESSkip x% Experts=50%, rho=0.482025.11 | 89.08 | — | — | — | — | — | — | — | |
| OURS (KG SELF-PLAY)PARAMS=4B2026.05 | 89.02 | — | — | — | — | — | — | — | |
| Qwen3-VL-8B + Qwen3-8B / H-OPDStudent=Qwen3-VL-4B-Instruct2026.07 | 88.9 | — | — | — | — | — | — | — | |
| EMOVAModel Size=72B2024.09 | 88.7 | — | — | — | — | — | — | — | |
| MoDESSkip x% Experts=67%, rho=0.652025.11 | 88.24 | — | — | — | — | — | — | — | |
| Phantom-7BParameters=7B2024.12 | 87.8 | — | — | — | — | — | — | — | |
| OpenVLThinkerV2Parameters=8B2026.04 | 87.4 | — | — | — | — | — | — | — | |
| Phantom-3.8BParameters=3.8B2024.12 | 87.3 | — | — | — | — | — | — | — | |
| VISEModel Scale=32B2026.06 | 87.27 | — | — | — | — | — | — | — | |
| InternVL-2.5-26BLatency=0.74542025.12 | 87.2 | — | — | — | — | — | — | — | |
| Intern-S1variant=mini, parameters=9B2026.01 | 87.12 | — | — | — | — | — | — | — | |
| VisionZero-CLEVRModel Scale=32B2026.06 | 87.02 | — | — | — | — | — | — | — | |
| Phantom-1.8BParameters=1.8B2024.12 | 87 | — | — | — | — | — | — | — | |
| VisionZero-ChartModel Scale=32B2026.06 | 86.94 | — | — | — | — | — | — | — | |
| iReasonerModel Scale=32B2026.06 | 86.9 | — | — | — | — | — | — | — | |
| InternVL3.5parameters=8B2026.01 | 86.88 | — | — | — | — | — | — | — | |
| EvoLMMModel Scale=32B2026.06 | 86.83 | — | — | — | — | — | — | — | |
| Innovator-VLvariant=8B-Instruct, parameters=8B2026.01 | 86.8 | — | — | — | — | — | — | — | |
| InternVL-3.5Model Size=8B2026.03 | 86.7 | — | — | — | — | — | — | — | |
| GPT-4o2026.04 | 86.7 | — | — | — | — | — | — | — | |
| LLaVA-OVversion=1.5, parameters=8B2026.01 | 86.64 | — | — | — | — | — | — | — | |
| Penguin-VLParameters=2B2026.03 | 86.6 | — | — | — | — | — | — | — | |
| InternVL2-26BModel=InternVL2-26B, Bitwidth=FP162026.03 | 86.44 | — | — | — | — | — | — | — | |
| VisionZero2026.04 | 86.3 | — | — | — | — | — | — | — | |
| QWEN3-VL + SFT ONLYPARAMS=4B2026.05 | 86.29 | — | — | — | — | — | — | — | |
| Innovator-VLvariant=8B-Thinking, parameters=8B2026.01 | 86.24 | — | — | — | — | — | — | — | |
| Qwen3-VL 8B-RModel Scale=8B, Strategy=Reasoning2026.04 | 86.2 | — | — | — | — | — | — | — | |
| VLsI-7BParameters=7B2024.12 | 86.1 | — | — | — | — | — | — | — | |
| VisionZero-RWModel Scale=32B2026.06 | 86.05 | — | — | — | — | — | — | — | |
| Qwen3-VL 8B + 4B-RMain Model=8B, Source Model=4B-R, Strategy=Reasoning Transfer2026.04 | 86 | — | — | — | — | — | — | — | |
| BaseModel Scale=32B2026.06 | 86 | — | — | — | — | — | — | — | |
| VisPlayModel Scale=32B2026.06 | 85.94 | — | — | — | — | — | — | — | |
| VLSI-2BParameters=2B2024.12 | 85.8 | — | — | — | — | — | — | — | |
| GPT-4oInference mode=Zero-shot2024.10 | 85.7 | — | — | — | — | — | — | — | |
| GPT-4o2024.09 | 85.7 | — | — | — | — | — | — | — | |
| OpenVLThinker-7BParameters=7B2026.04 | 85.7 | — | — | — | — | — | — | — | |
| D.P.Data Source Type=Real Data Reweighting, Sampling Strategy Category=Ours, Compute Budget=280K datapoints2026.03 | 85.6 | — | — | — | — | — | — | — | |
| Vision-G12026.04 | 85.6 | — | — | — | — | — | — | — | |
| VisionZero-ChartModel Scale=8B2026.06 | 85.49 | — | — | — | — | — | — | — | |
| VISEModel Scale=8B2026.06 | 85.41 | — | — | — | — | — | — | — | |
| InternVL-2.5-26B*Latency=0.74542025.12 | 85.4 | — | — | — | — | — | — | — | |
| Qwen3-VL 8B + 2B-RMain Model=8B, Source Model=2B-R, Strategy=Reasoning Transfer2026.04 | 85.3 | — | — | — | — | — | — | — | |
| QIGModel=InternVL2-26B, Bitwidth=W4A82026.03 | 85.24 | — | — | — | — | — | — | — | |
| VisionZero-CLEVRModel Scale=8B2026.06 | 85.18 | — | — | — | — | — | — | — | |
| LUVCBackbone=InternVL-2.5-26B*, Latency=0.4302, Speedup=1.732025.12 | 85.16 | — | — | — | — | — | — | — | |
| Qwen3-VLparameters=8B2026.01 | 85.12 | — | — | — | — | — | — | — | |
| QIGModel=InternVL2-26B, Bitwidth=W3A162026.03 | 85.12 | — | — | — | — | — | — | — | |
| iReasonerModel Scale=8B2026.06 | 85.12 | — | — | — | — | — | — | — | |
| FastVtoken ratio=64%2024.12 | 85.08 | — | 99.84 | — | — | — | — | — | |
| EvoLMMModel Scale=8B2026.06 | 85.05 | — | — | — | — | — | — | — | |
| SGPtoken ratio=64%2024.12 | 85.04 | — | 100.14 | — | — | — | — | — | |
| VISEModel Scale=4B2026.06 | 84.96 | — | — | — | — | — | — | — | |
| VisionZero-RWModel Scale=8B2026.06 | 84.94 | — | — | — | — | — | — | — | |
| InternVL2-26Btoken ratio=100%2024.12 | 84.92 | — | 100 | — | — | — | — | — | |
| EMOVAModel Size=7B2024.09 | 84.9 | — | — | — | — | — | — | — | |
| OracleData Source Type=Real Data Reweighting, Sampling Strategy Category=Gold, Compute Budget=280K datapoints2026.03 | 84.9 | — | — | — | — | — | — | — | |
| BaseModel Scale=8B2026.06 | 84.87 | — | — | — | — | — | — | — | |
| InternVL-2.5-8BLatency=0.29852025.12 | 84.8 | — | — | — | — | — | — | — | |
| ICONSData Source Type=Real Data Reweighting, Sampling Strategy Category=Baseline, Compute Budget=280K datapoints2026.03 | 84.8 | — | — | — | — | — | — | — | |
| NVILA-Lite-8B2026.05 | 84.8 | — | — | — | — | — | — | — | |
| VisPlayModel Scale=8B2026.06 | 84.79 | — | — | — | — | — | — | — | |
| Qwen3-VL 8B-SModel Scale=8B, Strategy=Simple2026.04 | 84.7 | — | — | — | — | — | — | — | |
| VL-Rethinker-7BParameters=7B2026.04 | 84.7 | — | — | — | — | — | — | — | |
| QWEN3-VL (VANILLA)PARAMS=4B2026.05 | 84.6 | — | — | — | — | — | — | — | |
| MBQModel=InternVL2-26B, Bitwidth=W3A162026.03 | 84.48 | — | — | — | — | — | — | — | |
| GR3D-8BTraining Stage=Stage 12026.05 | 84.48 | — | — | — | — | — | — | — | |
| MBQModel=InternVL2-26B, Bitwidth=W4A82026.03 | 84.44 | — | — | — | — | — | — | — | |
| Qwen3-VL-4BModel=Qwen3-VL, Parameter Count=4B2026.04 | 84.44 | — | — | — | — | — | — | — | |
| VisionZero-ChartModel Scale=4B2026.06 | 84.38 | — | — | — | — | — | — | — | |
| MoDESSkip x% Experts=83%, rho=0.82025.11 | 84.2 | — | — | — | — | — | — | — | |
| Qwen2.5-VL-3BModel=Qwen2.5-VL, Parameter Count=3B2026.04 | 84.12 | — | — | — | — | — | — | — | |
| Qwen2.5VL 3B# Vis tok.=14002025.04 | 84 | — | — | — | — | — | — | — | |
| GR3D-8BTraining Stage=Stage 22026.05 | 84 | — | — | — | — | — | — | — | |
| BaseRatio=100%, Backbone=Qwen2.5-VL-3B2026.05 | 83.96 | — | — | — | — | — | — | — | |
| iReasonerModel Scale=4B2026.06 | 83.72 | — | — | — | — | — | — | — | |
| UniformData Source Type=Real Data Reweighting, Sampling Strategy Category=Baseline, Compute Budget=280K datapoints2026.03 | 83.7 | — | — | — | — | — | — | — | |
| VisionZero-CLEVRModel Scale=4B2026.06 | 83.68 | — | — | — | — | — | — | — | |
| BalCapRL-3BBackbone=QwenVL2.5-3B2026.05 | 83.64 | — | — | — | — | — | — | — | |
| EvoLMMModel Scale=4B2026.06 | 83.52 | — | — | — | — | — | — | — | |
| FLARE-X 8B# Vis tok.=14002025.04 | 83.4 | — | — | — | — | — | — | — | |
| CapRL-3BBackbone=QwenVL2.5-3B2026.05 | 83.4 | — | — | — | — | — | — | — | |
| InternVL-2.5-8B*Latency=0.29852025.12 | 83.36 | — | — | — | — | — | — | — | |
| InternVL2-8BParameters=8B2024.12 | 83.3 | — | — | — | — | — | — | — | |
| Gemini 2.5 Pro2026.04 | 83.3 | — | — | — | — | — | — | — | |
| QwenVL2.5-3BModel size=3B2026.05 | 83.28 | — | — | — | — | — | — | — | |
| Qwen3-VL 4B-RModel Scale=4B, Strategy=Reasoning2026.04 | 83.2 | — | — | — | — | — | — | — | |
| BaseModel Scale=4B2026.06 | 83.18 | — | — | — | — | — | — | — | |
| BasePruning Ratio=100%, Base Model=Qwen3-VL-8B, Zero-shot evaluation=True2026.05 | 83.16 | — | — | — | — | — | — | — | |
| QWEN3-VL + SPIN (NO KG)PARAMS=4B2026.05 | 83.14 | — | — | — | — | — | — | — | |
| Qwen3-VL 4B-SModel Scale=4B, Strategy=Simple2026.04 | 83.1 | — | — | — | — | — | — | — | |
| Baseline (No Recompute)Model=Qwen3-VL-8B, Recomputation budget (k)=02026.03 | 83.08 | — | — | — | — | — | — | — |