multimodal benchmark
PerceptionBench Leaderboard
PerceptionBench is Moonshot AI's internal multimodal benchmark for evaluating atomic visual perception capabilities.
Updated Sep 5, 2026
Models2
Model coverage2
MetricScore
EvidenceB
PerceptionBench Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAC | Score63.50% | Percentile100.00% | Participants2 | EvidenceC | Evaluated |
| Rank02 | ModelMA | Score58.50% | Percentile0.00% | Participants2 | EvidenceC | Evaluated |
PerceptionBench Highlights
The leading models and scores on this benchmark.
PerceptionBench Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for PerceptionBench
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is PerceptionBench?
What PerceptionBench measures and how its scores work.
PerceptionBench is Moonshot AI's internal benchmark for evaluating atomic visual perception capabilities.
It measures atomic visual perception capabilities using the Score metric, reported as a ratio.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- PerceptionBench
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about PerceptionBench.
Which model scores highest on PerceptionBench?
Qwen3.8 Max is currently ranked first with 63.50%.
What are the top three models on PerceptionBench?
The current leaders are Qwen3.8 Max (63.50%) and Kimi K3 (58.50%).
Which PerceptionBench model has the lowest official input price?
Qwen3.8 Max has the lowest matched official input price at $2.0 input / $6.0 output per 1M tokens.
Which models are fastest among PerceptionBench results?
The fastest matched records are Qwen3.8 Max (60.87 tok/s via DeepInfra).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.