List your company
ResearchFormerly WizWand

Fine-grained Perception on HRBench 4K

86.8Pass@1Kimi K2.5 + ETCHR
50.50459.92769.3578.773May 22, 2026May 23, 2026May 24, 2026May 25, 2026May 26, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Kimi K2.5 + ETCHRTemperature=02026.0586.8
Kimi K2.5Temperature=02026.0585.6
BaseBackbone=Qwen3-VL-Thinking, Model Size=32B2026.0585.3
SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.0583.3
Gemini-3.1-Flash-Lite + ETCHRTemperature=02026.0582.9
Gemini-3.1-Flash-LiteTemperature=02026.0581.5
SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.0580.9
Qwen3-VL-8B + ETCHRTemperature=02026.0580.6
SFTBackbone=Qwen3-VL-Thinking, Model Size=4B2026.0579.5
SFTBackbone=Qwen3-VL-Thinking, Model Size=8B2026.0579.5
SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.0579.5
SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.0578.8
Qwen3-VL-8BTemperature=02026.0578.1
DeepEyesV2Temperature=0, Max tool invocations=102026.0577.9
SFT + AXPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.0577.9
ThymeTemperature=0, Max tool invocations=102026.0577.1
GRPOBackbone=Qwen3-VL-Thinking, Model Size=8B2026.0577.1
GRPOBackbone=Qwen3-VL-Thinking, Model Size=4B2026.0575.5
SFT + GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.0574.3
BaseBackbone=Qwen3-VL-Thinking, Model Size=4B2026.0573.6
SFTBackbone=Qwen3-VL-Thinking, Model Size=2B2026.0573.1
BaseBackbone=Qwen3-VL-Thinking, Model Size=8B2026.0572.8
GRPOBackbone=Qwen3-VL-Thinking, Model Size=2B2026.0565
BaseBackbone=Qwen3-VL-Thinking, Model Size=2B2026.0564.6
ThinkMorph-7BTemperature=0, Max image generations=102026.0559.6
Bagel-Zebra-CoTTemperature=0, Max image generations=102026.0551.9