# PerceptionBench

> Source: Benchize (https://benchize.com) — https://benchize.com/benchmarks/perceptionbench. When citing or reusing this content, credit "Benchize" and link to https://benchize.com/benchmarks/perceptionbench.

Atomic visual perception across the ten capabilities defined by PerceptionBench.

Canonical URL: https://benchize.com/benchmarks/perceptionbench
Last updated: Wed, 21 Oct 2026 18:05:10 GMT

Reproduction and quotation must name Benchize, link to the canonical HTML URL, and retain individual source citations.

Atomic visual perception across the ten capabilities defined by PerceptionBench.

Multimodal · Models tested: 19 · Score publishers: [moonshot k3](https://www.kimi.ai/blog/kimi-k3), perceptionbench

## Model results

| # | Model | Score | Source |  |
| --- | --- | --- | --- | --- |
| 1 / 5 | [GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol-max.md)Max | 59.7% | [Moonshot AI · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c](https://www.kimi.ai/blog/kimi-k3) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 59.7 \| percent \| reported \| Higher is better \| ### Configuration Reasoning effort max runs 3 thinkingEnabled true ### Reported statistics runCount 3 ### Source Score publisher [Moonshot AI](https://www.kimi.ai/blog/kimi-k3) Collected via [moonshot k3](https://www.kimi.ai/blog/kimi-k3) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://www.kimi.ai/blog/kimi-k3](https://www.kimi.ai/blog/kimi-k3) [Collected via](https://www.kimi.ai/blog/kimi-k3) Source Moonshot AI Original model label GPT 5.6 Sol (max) Source version sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c Observed 2026-10-21T16:31:44.130Z Evidence SHA-256 `0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c` Source locator `/flight/sheet/data/34/scores/2` |  |  |  |  |
| 1 / 14 | [GPT-5.6 Sol](https://benchize.com/models/gpt-5-6-sol.md) | 59.7% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 59.7 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 69.7 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 62.4 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 62.1 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 55.5 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 76.7 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 67.0 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 55.9 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 60.0 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 54.9 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 26.9 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label GPT-5.6-Sol Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/0` |  |  |  |  |
| 2 / 14 | [Kimi K3](https://benchize.com/models/kimi-k3.md) | 58.5% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 58.5 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 68.2 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 59.7 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 59.4 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 52.4 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 70.3 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 59.1 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 55.9 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 53.3 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 61.2 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 41.7 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Kimi K3 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/1` |  |  |  |  |
| 2 / 5 | [Kimi K3](https://benchize.com/models/kimi-k3-max.md)Max | 58.5% | [Moonshot AI · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c](https://www.kimi.ai/blog/kimi-k3) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 58.5 \| percent \| reported \| Higher is better \| ### Configuration Reasoning effort max runs 3 Temperature 1 thinkingEnabled true topP 1 ### Reported statistics runCount 3 ### Source Score publisher [Moonshot AI](https://www.kimi.ai/blog/kimi-k3) Collected via [moonshot k3](https://www.kimi.ai/blog/kimi-k3) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://www.kimi.ai/blog/kimi-k3](https://www.kimi.ai/blog/kimi-k3) [Collected via](https://www.kimi.ai/blog/kimi-k3) Source Moonshot AI Original model label Kimi K3 (max) Source version sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c Observed 2026-10-21T16:31:44.130Z Evidence SHA-256 `0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c` Source locator `/flight/sheet/data/34/scores/0` |  |  |  |  |
| 3 / 14 | [Claude Fable 5](https://benchize.com/models/claude-fable-5.md) | 57.2% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 57.2 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 58.5 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 52.9 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 60.9 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 51.5 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 70.4 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 56.1 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 51.6 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 59.8 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 64.3 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 45.0 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Claude-Fable-5 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/2` |  |  |  |  |
| 3 / 5 | [Claude Fable 5](https://benchize.com/models/claude-fable-5-max.md)Max | 57.2% | [Moonshot AI · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c](https://www.kimi.ai/blog/kimi-k3) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 57.2 \| percent \| reported \| Higher is better \| ### Configuration Fallback policy unspecified fallbackEnabled true Reasoning effort max runs 3 sourceFallbackLabel with fallback thinkingEnabled true Fallback is enabled for refusals; this does not mean every task used a fallback model. ### Reported statistics runCount 3 ### Source Score publisher [Moonshot AI](https://www.kimi.ai/blog/kimi-k3) Collected via [moonshot k3](https://www.kimi.ai/blog/kimi-k3) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://www.kimi.ai/blog/kimi-k3](https://www.kimi.ai/blog/kimi-k3) [Collected via](https://www.kimi.ai/blog/kimi-k3) Source Moonshot AI Original model label Claude Fable 5 (max, with fallback) Source version sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c Observed 2026-10-21T16:31:44.130Z Evidence SHA-256 `0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c` Source locator `/flight/sheet/data/34/scores/1` |  |  |  |  |
| 4 / 14 | [Gemini 3.1 Pro](https://benchize.com/models/gemini-3-1-pro.md) | 56.2% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 56.2 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 58.8 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 56.9 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 61.8 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 50.0 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 52.7 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 61.7 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 54.8 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 61.2 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 64.3 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 40.6 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Gemini-3.1-Pro Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/3` |  |  |  |  |
| 4 / 5 | [GPT-5.5](https://benchize.com/models/gpt-5-5-xhigh.md)Extra high | 55.8% | [Moonshot AI · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c](https://www.kimi.ai/blog/kimi-k3) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 55.8 \| percent \| reported \| Higher is better \| ### Configuration Reasoning effort xhigh runs 3 thinkingEnabled true ### Reported statistics runCount 3 ### Source Score publisher [Moonshot AI](https://www.kimi.ai/blog/kimi-k3) Collected via [moonshot k3](https://www.kimi.ai/blog/kimi-k3) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://www.kimi.ai/blog/kimi-k3](https://www.kimi.ai/blog/kimi-k3) [Collected via](https://www.kimi.ai/blog/kimi-k3) Source Moonshot AI Original model label GPT 5.5 (xhigh) Source version sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c Observed 2026-10-21T16:31:44.130Z Evidence SHA-256 `0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c` Source locator `/flight/sheet/data/34/scores/4` |  |  |  |  |
| 5 / 14 | [GPT-5.5](https://benchize.com/models/gpt-5-5.md) | 55.8% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 55.8 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 61.9 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 55.8 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 60.9 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 48.8 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 65.8 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 65.6 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 47.2 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 58.0 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 56.5 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 34.7 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label GPT-5.5 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/4` |  |  |  |  |
| 6 / 14 | [Gemini 3.5 Flash](https://benchize.com/models/gemini-3-5-flash.md) | 52% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 52.0 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 53.6 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 43.6 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 54.5 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 49.7 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 50.6 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 54.8 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 51.7 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 53.3 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 59.6 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 50.6 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Gemini-3.5-Flash Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/6` |  |  |  |  |
| 7 / 14 | [Qwen3.5 397B A17B](https://benchize.com/models/qwen3-5-397b-a17b.md) | 47.5% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 47.5 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 55.2 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 49.1 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 53.0 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 44.6 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 46.7 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 49.8 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 44.8 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 50.2 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 52.9 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 26.9 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Qwen3.5-397B-A17B Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/8` |  |  |  |  |
| 5 / 5 | [Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8-max.md)Max | 47.2% | [Moonshot AI · sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c](https://www.kimi.ai/blog/kimi-k3) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `score` \| 47.2 \| percent \| reported \| Higher is better \| ### Configuration Reasoning effort max runs 3 thinkingEnabled true ### Reported statistics runCount 3 ### Source Score publisher [Moonshot AI](https://www.kimi.ai/blog/kimi-k3) Collected via [moonshot k3](https://www.kimi.ai/blog/kimi-k3) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://www.kimi.ai/blog/kimi-k3](https://www.kimi.ai/blog/kimi-k3) [Collected via](https://www.kimi.ai/blog/kimi-k3) Source Moonshot AI Original model label Claude Opus 4.8 (max) Source version sha256:0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c Observed 2026-10-21T16:31:44.130Z Evidence SHA-256 `0e806bbad76dc8e8dbc899a86051e3406bdd5995fe6c6aa1b2176dde4fb28e6c` Source locator `/flight/sheet/data/34/scores/3` |  |  |  |  |
| 8 / 14 | [Claude Opus 4.8](https://benchize.com/models/claude-opus-4-8.md) | 47.2% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 47.2 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 51.4 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 44.2 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 49.4 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 40.6 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 58.8 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 48.4 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 40.7 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 44.7 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 54.1 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 38.7 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Claude-Opus-4.8 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/9` |  |  |  |  |
| 9 / 14 | [Kimi K2.6](https://benchize.com/models/kimi-k2-6.md) | 42.6% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 42.6 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 50.9 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 45.2 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 43.6 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 42.4 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 45.2 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 39.1 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 34.5 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 40.4 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 40.8 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 41.0 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Kimi K2.6 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/10` |  |  |  |  |
| 10 / 14 | [Grok 4.5](https://benchize.com/models/grok-4-5.md) | 41% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 41.0 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 47.0 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 35.2 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 39.4 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 41.2 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 39.7 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 43.7 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 36.2 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 39.6 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 43.9 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 44.7 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Grok-4.5 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/11` |  |  |  |  |
| 11 / 14 | [Gemma 4 31B](https://benchize.com/models/gemma-4-31b-it.md) | 40.7% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 40.7 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 42.7 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 33.9 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 40.3 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 39.1 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 44.9 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 43.7 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 39.0 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 45.9 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 46.7 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 32.1 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Gemma-4-31B Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/12` |  |  |  |  |
| 12 / 14 | [GLM-5V Turbo](https://benchize.com/models/glm-5v-turbo.md) | 39.6% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 39.6 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 41.2 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 40.0 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 41.2 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 41.2 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 43.9 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 45.2 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 36.6 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 32.9 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 43.5 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 28.0 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label GLM-5V-Turbo Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/13` |  |  |  |  |
| 13 / 14 | [MiniMax M3](https://benchize.com/models/minimax-m3.md) | 33.1% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 33.1 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 40.0 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 30.3 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 34.6 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 36.7 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 33.3 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 31.2 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 26.6 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 31.0 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 35.7 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 29.9 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label Minimax-M3 Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/14` |  |  |  |  |
| 14 / 14 | [GLM-4.6V](https://benchize.com/models/glm-4-6v.md) | 32.5% | [Moonshot AI / PerceptionBench · ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://github.com/MoonshotAI/PerceptionBench) |  |
| ### Original source values \| Metric \| Value \| Unit \| Scope \| Direction \| \| --- \| --- \| --- \| --- \| --- \| \| `accuracy` \| 32.5 \| percent \| overall \| Higher is better \| \| `visual_relation_accuracy` \| 35.2 \| percent \| atomic_capability \| Higher is better \| \| `counting_accuracy` \| 31.8 \| percent \| atomic_capability \| Higher is better \| \| `attribute_accuracy` \| 35.2 \| percent \| atomic_capability \| Higher is better \| \| `depth_3d_accuracy` \| 29.1 \| percent \| atomic_capability \| Higher is better \| \| `localization_accuracy` \| 30.6 \| percent \| atomic_capability \| Higher is better \| \| `comparison_accuracy` \| 34.8 \| percent \| atomic_capability \| Higher is better \| \| `fine_grained_recognition_accuracy` \| 29.3 \| percent \| atomic_capability \| Higher is better \| \| `contextual_integration_accuracy` \| 33.7 \| percent \| atomic_capability \| Higher is better \| \| `ocr_accuracy` \| 39.2 \| percent \| atomic_capability \| Higher is better \| \| `hallucination_accuracy` \| 26.9 \| percent \| atomic_capability \| Higher is better \| ### Configuration judgeModel gpt-oss-120b promptProtocol unified reasoningBudgetPolicy highest_available ### Reported statistics sampleSize 3000 ### Source Score publisher [Moonshot AI / PerceptionBench](https://github.com/MoonshotAI/PerceptionBench) Collected via [perceptionbench](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Publisher reference Referenced by the collection source; the upstream result has not been independently captured. Benchmark [https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) [Collected via](https://api.github.com/repos/MoonshotAI/PerceptionBench/contents/README.md?ref=ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e) Source Moonshot AI / PerceptionBench Original model label GLM-4.6V Source version ba032c06e9b6ee3679171ff6ba643b7a0cfebe2e Observed 2026-10-21T16:33:36.858Z Evidence SHA-256 `0863d20bc397e2324ea76099fe703aed419dc7f0dbb359e522f7dac07dda6529` Source locator `/leaderboard/rows/15` |  |  |  |  |

---
Cite as: PerceptionBench — Benchize, https://benchize.com/benchmarks/perceptionbench (retrieved 2026-10-21)