Which model leads TextVQA?

Qwen2-VL-72B-Instruct leads the TextVQA board of 12 measured models on the Vector Wire tracker at 85.5. Every score is source-attributed with a verification tier.

TextVQA Leaderboard — Model Scores

Refresh Models —Deals —Regulatory —Sources —

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

The field · 12 measured

57.885.5

12 models measured, most on the official card harness. Qwen2-VL-72B-Instruct tops the board at 85.5.

12 measured·12 new this week·Lifecycle Updated Sep 4, 2026Basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 12Bars show 57.8–85.5· ◆ solid = first-party or better · outlined = arm's-length

01Qwen2-VL-72B-InstructNEWAlibaba85.5———

02Qwen2.5-Omni-7BNEWAlibaba84.4$0.10/M$6.8/M$0.041

03LFM2.5-VL-3BNEWLiquid AI84.3———

04Qwen3-VL-4BNEWAlibaba81.83P———

05Nova ProNEWAmazon81.5$0.80/M$3.2/M$0.025

06Nova LiteNEWAmazon80.2$0.06/M$0.24/M$0.002

07Qwen3-VL-2BNEWAlibaba79.93P———

08Phi-4-multimodal-instructNEWMicrosoft75.6———

09Phi-3.5-vision-instructNEWMicrosoft72———

10Gemma 3 12BNEWGoogle67.7$0.05/M$0.15/M$0.001

11Gemma 3 27BNEWGoogle65.1$0.08/M$0.16/M$0.002

12Gemma 3 4BNEWGoogle57.8$0.05/M$0.10/M$0.001

◆ 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.