LongShOTBench: leaderboard

MBZUAI LongShOTBench (arXiv:2512.16978): long-form omni-modal video understanding benchmark combining visual, speech, and ambient-audio cues across open-ended single- and multi-turn questions graded with weighted criterion-level rubrics (0-100%). Overall score across standalone models (multi-model agentic frameworks excluded); higher is better.

Metric: Weighted Rubric Score (%). Source: longshot.cvmbzuai.com. Saturation forecast: Around 2028. 102 models tracked. Updated 21 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3 Omni 30B A3B (Thinking)64.2
2Gemini 3.1 Pro (Preview)56.97#99
3Voxtral-Mini-3B-250749.61#3682
4Voxtral-Small-24B-250748.56#1246
5Qwen 3 VL 235B A22B (Thinking)47.59#459
6Qwen 3 VL 32B (Thinking)46.72#573
7Vero-Qwen3I-8B46.18
8Kimi K2.645.07#192
9Qwen 3 VL 30B A3B (Thinking)43.71#719
10Qwen 3 VL 235B A22B Instruct42.58#532
11Qwen 3 VL 8B (Thinking)42.18#910
12Gemma 3 27B (IT)41.97#1324
13Intern-S1-mini41.45#1178
14Nemotron 3 Nano Omni 30B A3BReasoning41.22#1412
15Qwen 3 VL 32B Instruct40.59#606

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=longshotbench · How It Works · Data refreshed daily, snapshot 2026-11-21.