Benchmark Almanac

3DSceneBench

GPT-6 Astra 59.5 leads since November 2026
Not in model cards

Agents reconstruct interactive MuJoCo scenes from images and task descriptions. Verifiers measure appearance, geometry, and physical functionality, then average the three scores.

The full suite contains 49 scenes. The public mini package contains seven scenes. Full-suite leaderboard scores must not be presented as mini scores.

Released
November 2026
Kind
Physical prediction
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text, Images → 3D, Actions
Interaction
Agentic
Environment
MuJoCo
Items from
Generated
Refresh
One fixed set
Run by
A host
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
10
Models scored on its own leaderboard

GPT-6 Astra leads with 59.5

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.

  1. 1 GPT-6 Astra L0; max effort 59.5
  2. 2 Gemini 4 Argon L0; high effort 56.9
  3. 3 Claude Opus 5.5 L0; max effort 56.8
  4. 4 Kimi K3.1 Pro L0; high effort 37.1
  5. 5 DeepSeek V4.1 Flash L0; high effort 30.6
  6. 6 MiniMax-M3.1-Flash L0; high effort 28.8
  7. 7 Qwen 3.8 Max L0; high effort 26.1
  8. 8 MiMo-V2.6 Pro L0; high effort 25.0
  9. 9 GLM-5.3-Flash L0; high effort 24.5
  10. 10 Mistral Large 4 L0; high effort 24.0

Source: Official 3DSceneBench leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 59.5 L0; max effort Sep 2026 Official leaderboard
2 Gemini 4 Argon 56.9 L0; high effort Sep 2026 Official leaderboard
3 Claude Opus 5.5 56.8 L0; max effort Sep 2026 Official leaderboard
4 Kimi K3.1 Pro 37.1 L0; high effort Oct 2026 Official leaderboard
5 DeepSeek V4.1 Flash 30.6 L0; high effort Sep 2026 Official leaderboard
6 MiniMax-M3.1-Flash 28.8 L0; high effort Oct 2026 Official leaderboard
7 Qwen 3.8 Max 26.1 L0; high effort Aug 2026 Official leaderboard
8 MiMo-V2.6 Pro 25.0 L0; high effort Sep 2026 Official leaderboard
9 GLM-5.3-Flash 24.5 L0; high effort Aug 2026 Official leaderboard
10 Mistral Large 4 24.0 L0; high effort Oct 2026 Official leaderboard

Rankings by part

We read 12 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Appearance, Geometry and Functionality. Gemini 4 Argon leads 7 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Appearance

10 models we track have a score here.

  1. 1 GPT-6 Astra 79.3
  2. 2 Claude Opus 5.5 ↑1 70.5
  3. 3 Gemini 4 Argon ↓1 66.7
  4. 4 Kimi K3.1 Pro 55.3
  5. 5 MiniMax-M3.1-Flash ↑1 48.2
  6. 6 DeepSeek V4.1 Flash ↓1 45.1
  7. 7 MiMo-V2.6 Pro ↑1 40.0
  8. 8 Qwen 3.8 Max ↓1 39.0
  9. 9 GLM-5.3-Flash 36.9
  10. 10 Mistral Large 4 36.2

Number one over time

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

OpenAI
Own leaderboard 1 leader

Source: Official 3DSceneBench leaderboard

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
On the benchmark’s own leaderboard
20.0 30.0 40.0 50.0 60.0 2027 20.0 30.0 40.0 50.0 60.0 2027

Source: Official 3DSceneBench 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 World models

All 53 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. GPT-5.4 58.3
  • 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%
  • VPCT 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. Gemini 3 Pro 91%
  • 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%
  • RBench Only in system 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%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (10 entries).
  • Pass Models scored 10 models we track.
  • Pass Newest score The newest is from October 2026, for Kimi K3.1 Pro.
  • Pass Frontier models 3 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on pranalabs.github.io: 12 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