Benchmark

ZEROBench

ZEROBench is a challenging vision benchmark designed to test models on zero-shot visual understanding tasks.

Modality
image
Categories
multimodal, reasoning, vision
Openness
unknown
Source
llm_stats
Reported scores
9

Reported scores

llm_stats

ModelOrganizationReported valueReportedEvidence
Kimi K2.5Moonshot AI0.112026-01-27evidence
Kimi K3Moonshot AI0.412026-07-16evidence
Muse SparkMeta0.332026-04-08evidence
Qwen3 VL 235B A22B ThinkingQwen0.042025-09-22evidence
Qwen3.5-122B-A10BQwen0.092026-02-24evidence
Qwen3.5-27BQwen0.12026-02-24evidence
Qwen3.5-35B-A3BQwen0.082026-02-24evidence
Seed 2.1 ProByteDance0.182026-06-24evidence
Seed 2.1 TurboByteDance0.112026-06-24evidence

Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.