Benchmark Almanac

MineAmongUs

Gemini 3 Flash 81.3% leads since August 2026
Not in model cards

Vision-language agents play an embodied Among Us game with verbal and nonverbal deception.

The August 2026 paper compares multiple VLM agents. The abstract does not state the best score.

Released
August 2026
Kind
Multi-agent games
Level
Everyday
Scored
Settled by events
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Images → Actions, Text
Interaction
Multi-agent
Environment
Among Us
Items from
Live play or votes
Refresh
Live
Run by
Anyone
Strongest model tested
Gemini 3 Flash #60 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
12
Models scored in its paper

Gemini 3 Flash leads with 81.3% among models released by April 2026

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The headline is the mean of imposter and crewmate win rates against the same twelve opponents. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 Gemini 3 Flash Case1B 81.3%
  2. 2 Gemini 3.1 Flash-Lite Case2A 77.1%
  3. 3 GPT-4.1 mini Case2A 70.8%
  4. 4 GPT-5 mini Case2A 70.8%
  5. 5 Kimi K2.5 Case1B 64.6%
  6. 6 Gemma 4 31B IT Case1B 62.5%
  7. 7 Qwen3.5-27B Case2B 58.3%
  8. 8 Qwen3.6 27B Case2B 56.3%
  9. 9 Gemma 4 26B A4B Case1B 47.9%
  10. 10 GPT-5 Case2A 43.8%
  11. 11 Gemini 2.5 Flash Sep 2025, Case2A 37.5%
  12. 12 Qwen3.5-9B Case1B 22.9%

Source: The MineAmongUs paper

#ModelScoreSettingDateSource
1 Gemini 3 Flash 81.3% Case1B, ARIA Dec 2025 Official leaderboard
2 Gemini 3.1 Flash-Lite 77.1% Case2A, ARIA Mar 2026 Official leaderboard
3 GPT-4.1 mini 70.8% Case2A, ARIA Apr 2025 Official leaderboard
4 GPT-5 mini 70.8% Case2A, ARIA Aug 2025 Official leaderboard
5 Kimi K2.5 64.6% Case1B, ARIA Jan 2026 Official leaderboard
6 Gemma 4 31B IT 62.5% Case1B, ARIA Apr 2026 Official leaderboard
7 Qwen3.5-27B 58.3% Case2B, ARIA Feb 2026 Official leaderboard
8 Qwen3.6 27B 56.3% Case2B, ARIA Apr 2026 Official leaderboard
9 Gemma 4 26B A4B 47.9% Case1B, ARIA Apr 2026 Official leaderboard
10 GPT-5 43.8% Case2A, ARIA Aug 2025 Official leaderboard
11 Gemini 2.5 Flash Sep 2025 37.5% Case2A, ARIA Sep 2025 Official leaderboard
12 Qwen3.5-9B 22.9% Case1B, ARIA Feb 2026 Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s paper: Imposter win rate and Crewmate win rate. Each ranks models on one part of the test. Gemini 3 Flash leads both. Arrows show how far a model moved from its overall rank among the models we track.

Imposter win rate

12 models we track have a score here.

  1. 1 Gemini 3 Flash 83.3%
  2. 2 Kimi K2.5 ↑3 70.8%
  3. 3 Qwen3.5-27B ↑4 70.8%
  4. 4 Gemini 3.1 Flash-Lite ↓2 70.8%
  5. 5 GPT-4.1 mini ↓2 66.7%
  6. 6 GPT-5 ↑4 66.7%
  7. 7 GPT-5 mini ↓3 66.7%
  8. 8 Qwen3.6 27B 62.5%
  9. 9 Gemma 4 31B IT ↓3 58.3%
  10. 10 Gemini 2.5 Flash Sep 2025 ↑1 54.2%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Gemini 3 Flash ARIA
Case2A Case1A
81.3%
Gemini 3.1 Flash-Lite ARIA
Case2B Case1A
77.1%
GPT-4.1 mini ARIA
Case2B Case1B Case2A
70.8%
GPT-5 mini ARIA
Case1B Case1A Case2B Case2A
70.8%
Kimi K2.5 ARIA
Case1A Case1B
64.6%
Gemma 4 31B IT ARIA
Case2B Case2A Case1A
62.5%
Qwen3.5-27B ARIA
Case1B Case2A Case2B
58.3%
Qwen3.6 27B ARIA
Case2A Case1A
56.3%
Gemma 4 26B A4B ARIA
Case2A Case1A
47.9%
GPT-5 ARIA
Case2B
43.8%
Gemini 2.5 Flash ARIA
Case1A
37.5%
Qwen3.5-9B ARIA
Case1A
22.9%

Source: The MineAmongUs paper

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 3 Flash Case1B ARIA 81.3%
Gemini 3 Flash Case2B ARIA 81.3%
Gemini 3 Flash Case1A ARIA 79.2%
Gemini 3.1 Flash-Lite Case2A ARIA 77.1%
Gemini 3.1 Flash-Lite Case1A ARIA 75%
GPT-4.1 mini Case2A ARIA 70.8%
GPT-5 mini Case2A ARIA 70.8%
Gemini 3.1 Flash-Lite Case1B ARIA 68.8%
Gemini 3.1 Flash-Lite Case2B ARIA 66.7%
Gemini 3 Flash Case2A ARIA 64.6%
Kimi K2.5 Case1B ARIA 64.6%
GPT-4.1 mini Case1B ARIA 62.5%
Kimi K2.5 Case2A ARIA 62.5%
Kimi K2.5 Case2B ARIA 62.5%
Gemma 4 31B IT Case1B ARIA 62.5%
Gemma 4 31B IT Case1A ARIA 60.4%
GPT-4.1 mini Case1A ARIA 58.3%
Qwen3.5-27B Case2B ARIA 58.3%
GPT-5 mini Case2B ARIA 56.3%
Kimi K2.5 Case1A ARIA 56.3%
Qwen3.6 27B Case2B ARIA 56.3%
GPT-4.1 mini Case2B ARIA 54.2%
Qwen3.6 27B Case1A ARIA 54.2%
Qwen3.6 27B Case1B ARIA 54.2%
Gemma 4 26B A4B Case1B ARIA 47.9%
Qwen3.5-27B Case1A ARIA 45.8%
Gemma 4 26B A4B Case1A ARIA 45.8%
GPT-5 Case2A ARIA 43.8%
Qwen3.5-27B Case2A ARIA 43.8%
GPT-5 Case1A ARIA 39.6%
GPT-5 Case1B ARIA 39.6%
Gemma 4 26B A4B Case2B ARIA 39.6%
Gemma 4 31B IT Case2A ARIA 39.6%
GPT-5 Case2B ARIA 37.5%
Gemini 2.5 Flash Sep 2025 Case2A ARIA 37.5%
Gemini 2.5 Flash Sep 2025 Case2B ARIA 37.5%
Gemini 2.5 Flash Sep 2025 Case1B ARIA 35.4%
Qwen3.5-27B Case1B ARIA 35.4%
Gemma 4 26B A4B Case2A ARIA 35.4%
Qwen3.6 27B Case2A ARIA 35.4%
GPT-5 mini Case1A ARIA 33.3%
Gemini 2.5 Flash Sep 2025 Case1A ARIA 33.3%
Gemma 4 31B IT Case2B ARIA 27.1%
GPT-5 mini Case1B ARIA 25%
Qwen3.5-9B Case1B ARIA 22.9%
Qwen3.5-9B Case2B ARIA 22.9%
Qwen3.5-9B Case1A ARIA 18.8%
Qwen3.5-9B Case2A ARIA 18.8%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Benchmark paper Not tested since Aug 2026

Source: The MineAmongUs paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Flash GPT-4.1 mini 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Gemini 3 Flash GPT-4.1 mini

Source: Official MineAmongUs leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Games and long tasks

All 261 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6.2 Astra 100%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Qwen3.6 27B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (12 entries).
  • Pass Models scored 12 models we track.
  • Warning Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Gemini 3 Flash (#60).
  • Pass Headroom The best is 81.3%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 48 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check