Which model leads DocVQA-val?

Q: Which model leads DocVQA-val?

Qwen2.5 VL 72B tops the board at 95.75

DocVQA-val Leaderboard — 19 Models Ranked

Pro Mode — raw scores, one benchmark under the lensofficial card harness

The field · 19 measured

76.295.75

19 models measured, most on the official card harness. Qwen2.5 VL 72B tops the board at 95.75.

19 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 19Bars show 76.2–95.75 · ◆ solid = first-party or better · outlined = arm's-length

01Qwen2.5 VL 72B+1 altAlibaba95.753P———

02Mage-VL-4BMicrosoft95.143P———

03Qwen3.5 4BAlibaba94.8$0.03/M$0.15/M$0.001

04Qwen3 VL 4B (Reasoning)Alibaba94.693P———

05Phi 4 Multimodal InstructMicrosoft92.793P———

06Qwen2.5 VL 3BAlibaba92.713P———

07Qwen3.5 2B+2 altsAlibaba92.6———

08North-Micro-Vision-Instruct+1 altCohere92.1———

09InternVL3_5-4BOpenGVLab91.8———

10LFM2.5-VL-3BLiquid AI91.1———

11LFM2-VL-3BLiquid AI89.8———

12Ministral 3 3B+1 altMistral89.6$0.10/M$0.10/M$0.001

13InternVL3_5-2BOpenGVLab88.4———

14LFM2.5-VL-1.6B+1 altLiquid AI87.7———

15Gemma 4 E4BGoogle87.4$0.02/M$0.10/M$0.001

16Phi 3.5 Vision Instruct+1 altMicrosoft86———

17Gemma 4 E2B+2 altsGoogle85.7———

18Qwen3 VL 2B Instruct+1 altAlibaba82.5———

19Phi 4 R V 15BMicrosoft76.23P———

◆ solid = independently verified · aggregator attested · vendor attributed — outlined = cross-reference · source-attributed. Dimmed ◆ = confidence inferred. “+N alts” = alternate disclosures in the drawer with why-demoted. Prices per million tokens; $/point = input price ÷ score. Every row opens its model dossier.

Verification: 19 models scored · 0 independently verified · 3 vendor cross-reference · 16 vendor-reported · 1 with source disagreement. How these tiers are assigned