Benchmark Almanac

VPCT

Gemini 3 Pro 91% leads since Nov 2025 Fading

Models view an image of a ball, ramps, and three buckets and predict which bucket the ball falls into. Scored by accuracy on 100 problems.

Made by Chase Brower Released Jan 2025 Physical prediction All tasks public Last in a launch post Dec 2025 , dropped after it saturated Last model added Mar 2026 Scores refreshed 31 Oct 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Multiple choice
camelCase12/vpct-runner 10 stars Dataset Leaderboard
28
Models scored in Epoch AI data
1
Launch post lists it, from 1 lab
2
System cards report it, from 1 lab

Gemini 3 Pro leads with 91% among models released by Dec 2025

Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in Dec 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

28 models. Scroll sideways for the rest.

  1. 91.0 Gemini 3 Pro
  2. 84.0 GPT-5.2
  3. 72.6 Gemini 3 Flash
  4. 66.0 GPT-5
  5. 58.7 GPT-5.1
  6. 57.5 o4-mini
  7. 52.0 o3
  8. 48.0 Gemini 2.5 Pro
  9. 46.4 Gemini 2.5 Pro
  10. 46.2 Gemini 2.5 Flash
  11. 45.0 GPT-4.5
  12. 40.8 gemini-robotics-er-1.5-preview
  13. 40.5 Gemini 2.5 Pro
  14. 40.2 GPT-5 mini
  15. 40.0 GPT-4o
  16. 40.0 Claude Opus 4.5
  17. 39.8 Claude Sonnet 4.5
  18. 39.0 Claude 3.7 Sonnet
  19. 38.0 Gemini 2.5 Flash
  20. 38.0 Claude Opus 4
  21. 37.2 GPT-5 nano
  22. 37.0 o1
  23. 35.0 Claude Opus 4.1
  24. 34.0 GPT-4o mini
  25. 34.0 Claude Sonnet 4
  26. 33.0 Claude 3.5 Sonnet
  27. 33.0 Claude 3.5 Sonnet (October 2024)
  28. 30.0 Gemini 2.5 Flash-Lite
  1. 1 Gemini 3 Pro 91%
  2. 2 GPT-5.2 xhigh 84%
  3. 3 Gemini 3 Flash 72.6%
  4. 4 GPT-5 high 66%
  5. 5 GPT-5.1 high 58.7%
  6. 6 o4-mini medium 57.5%
  7. 7 o3 medium 52%
  8. 8 Gemini 2.5 Pro Mar 2025 48%
  9. 9 Gemini 2.5 Pro Jun 2025 46.4%
  10. 10 Gemini 2.5 Flash Sep 2025 46.2%
  11. 11 GPT-4.5 45%
  12. 12 gemini-robotics-er-1.5-preview 40.8%
  13. 13 Gemini 2.5 Pro May 2025 40.5%
  14. 14 GPT-5 mini medium 40.2%
  15. 15 GPT-4o Nov 2024 40%
  16. 16 Claude Opus 4.5 32k thinking 40%
  17. 17 Claude Sonnet 4.5 32k thinking 39.8%
  18. 18 Claude 3.7 Sonnet 39%
  19. 19 Gemini 2.5 Flash Apr 2025 38%
  20. 20 Claude Opus 4 16K 38%
  21. 21 GPT-5 nano high 37.2%
  22. 22 o1 medium 37%
  23. 23 Claude Opus 4.1 35%
  24. 24 GPT-4o mini 34%
  25. 25 Claude Sonnet 4 32K 34%
  26. 26 Claude 3.5 Sonnet Jun 2024 33%
  27. 27 Claude 3.5 Sonnet (October 2024) 33%
  28. 28 Gemini 2.5 Flash-Lite Sep 2025 30%
#ModelScoreSettingDateSource
1 Gemini 3 Pro 91% Nov 2025 Epoch AI
2 GPT-5.2 84% xhigh Dec 2025 Epoch AI
3 Gemini 3 Flash 72.6% Dec 2025 Epoch AI
4 GPT-5 66% high Aug 2025 Epoch AI
5 GPT-5.1 58.7% high Nov 2025 Epoch AI
6 o4-mini 57.5% medium Apr 2025 Epoch AI
7 o3 52% medium Apr 2025 Epoch AI
8 Gemini 2.5 Pro Mar 2025 48% Mar 2025 Epoch AI
9 Gemini 2.5 Pro Jun 2025 46.4% Jun 2025 Epoch AI
10 Gemini 2.5 Flash Sep 2025 46.2% Sep 2025 Epoch AI
11 GPT-4.5 45% Feb 2025 Epoch AI
12 gemini-robotics-er-1.5-preview 40.8% Sep 2025 Epoch AI
13 Gemini 2.5 Pro May 2025 40.5% May 2025 Epoch AI
14 GPT-5 mini 40.2% medium Aug 2025 Epoch AI
15 GPT-4o Nov 2024 40% Nov 2024 Epoch AI
16 Claude Opus 4.5 40% 32k thinking Nov 2025 Epoch AI
17 Claude Sonnet 4.5 39.8% 32k thinking Sep 2025 Epoch AI
18 Claude 3.7 Sonnet 39% Feb 2025 Epoch AI
19 Gemini 2.5 Flash Apr 2025 38% Apr 2025 Epoch AI
20 Claude Opus 4 38% 16K May 2025 Epoch AI
21 GPT-5 nano 37.2% high Aug 2025 Epoch AI
22 o1 37% medium Dec 2024 Epoch AI
23 Claude Opus 4.1 35% Aug 2025 Epoch AI
24 GPT-4o mini 34% Jul 2024 Epoch AI
25 Claude Sonnet 4 34% 32K May 2025 Epoch AI
26 Claude 3.5 Sonnet 33% Jun 2024 Jun 2024 Epoch AI
27 Claude 3.5 Sonnet (October 2024) 33% Oct 2024 Epoch AI
28 Gemini 2.5 Flash-Lite Sep 2025 30% Sep 2025 Epoch AI

Labs report scores for 2 models in their cards. We plot every source in the charts below.

6 models have held number one since its release in Jan 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for Epoch AI’s runs and what labs claim in their cards.

OpenAI Google DeepMind ByteDance Seed
Epoch AI runs Passed 85% in Nov 2025, by Gemini 3 Pro
Gemini 3 Pro 91%
Lab claims Not tested since Mar 2026
Seed2.0 Pro 76%

The best score passed 85% in Nov 2025

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

Run or collected by Epoch AI Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Random guessing Gemini 3 Pro GPT-5 o4-mini Gemini 2.5 Pro GPT-4o GPT-4o mini 0% 20% 40% 60% 80% 100% 2024 2025 2026 Near the ceiling (85%) Random guessing Gemini 3 Pro GPT-5

Labs headline it less often now, last in Dec 2025

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

ByteDance Seed
DateLaunch postScoreAs the post writes it
18 Dec 2025 Seed1.8 61 VPCT

2 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
21 Mar 2026 Seed1.8 model card 61 VPCT (Table 2) Capabilities
14 Feb 2026 Seed2.0 Pro model card 76 VPCT (Table 8) Capabilities

More in World models

All 34 in World models
  • AgentWorldBench In use Language world models predict the next environment observation after an agent action, in 7 domains from MCP and search to Android and desktop OS. A judge model scores each prediction against the real observation on five dimensions. Qwen-AgentWorld-35B-A3B 56.4
  • IntPhys 2 In use Models watch simulated videos and judge whether each event is physically possible, testing object permanence, immutability, continuity and solidity. Scored by accuracy. Muse Spark Open 68.3%
  • QwenWorldBench Fading Models act as world models that simulate agent environments in 7 domains, such as terminal, web, and Android. A rubric judge scores five dimensions. Qwen3.7-Plus 62.1%
  • ContPhy Fading Models infer physical properties such as mass and density of fluids and soft bodies from videos and predict their dynamics. Seed2.0 Pro 67.4%
  • MVPBench Dropped Models answer video questions on spatial, temporal and intuitive physics. Each question comes in a pair of near-identical videos with opposite answers, so text or single-frame shortcuts fail. gemini-robotics-er-1.5-preview 72.5%
  • RBench Not in model cards 650 image+text prompts for robot video generation (tasks and embodiments). MLLM judge scores task completion, physical plausibility, stability. Wan 2.6 60.7%