Which model leads RefSpatial-Bench?

Q: Which model leads RefSpatial-Bench?

Qwen3.6 27B tops the board at 70

RefSpatial-Bench Leaderboard — 12 Models

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

The field · 11 measured

28.970

11 models measured, most on the official card harness. Qwen3.6 27B tops the board at 70.

11 measured·Lifecycle Updated Oct 8, 2026Score basis

#◆Vendor$/M in$/M out

Leader — ignites whiteField — fill cools with scoreIndependently verified only0 of 11Bars show 28.9–70 · ◆ solid = first-party or better · outlined = arm's-length

01Qwen3.6 27BAlibaba70$0.60/M$3.6/M$0.030

02Qwen3 VL 235B A22B ReasoningAlibaba69.9$0.40/M$4.0/M$0.031

03Qwen3.5 122B A10BAlibaba69.3$0.40/M$3.2/M$0.026

04Qwen3.5 27BAlibaba67.7$0.30/M$2.4/M$0.020

05Qwen3.6 35B A3BAlibaba64.3$0.38/M$2.3/M$0.020

06Qwen3.5 35B A3BAlibaba63.5$0.25/M$2.0/M$0.018

07Seed 1.8ByteDance56.3$0.25/M$2.0/M$0.020

08MiMo Embodied 7BXiaomi48———

09HY-Embodied 0.5 MoT-2BTencent45.8———

10Qwen3 VL 4B (Reasoning)Alibaba45.3———

11Qwen3 VL 2BAlibaba28.9———

◆ 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: 11 models scored · 0 independently verified · 3 vendor cross-reference · 8 vendor-reported · 0 with source disagreement. How these tiers are assigned