Benchmark Almanac

ReSpaceBench

Qwen3-VL-8B-SFT 60.4% tops its leaderboard GPT-5.6 Sol 50.5% leads the models we track
Not in model cards

Models recognize rooms across separate visits and track objects that move between visits. The authors test 32 models on synthetic and real homes.

The benchmark contains 3,244 questions across 590 scenes.

Released
November 2026
Kind
Embodied reasoning
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Measures
Capability
Language
English
In and out
Video, Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
A host
In views
Vision
Strongest model tested
GPT-5.6 Sol #14 on the capability index
Last model added
August 2026
Scores refreshed
November 20th, 2026
33
Entries in its paper, 17 models we track

Qwen3-VL-8B-SFT tops the leaderboard with 60.4% GPT-5.6 Sol leads with 50.5% among models released by August 2026

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is GPT-5.6 Sol, at 50.5%. This view shows only models we track, each at its best.

The newest model with a score here came out in August 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 Qwen3-VL-8B-SFT Video 60.4%
  2. 2 GPT-5.6 Sol Video 50.5%
  3. 3 Seed2.0 Pro Video 49.3%
  4. 4 Seed2.1 Pro Video 48%
  5. 5 Qwen 3.8 Max Video 46.3%
  6. 6 Qwen3-VL-32B Video 41.3%
  7. 7 Qwen3.6 27B Video 40.3%
  8. 8 InternVL3.5-38B Video 38%
  9. 9 Qwen 3.6 35B-A3B Video 36.8%
  10. 10 Mage-VL Video 35.7%
  11. 11 Qwen3-VL-8B-Instruct Video 35.5%
  12. 12 LLaVA-OneVision-2-8B Video 34.4%
  13. 13 Gemini 3.1 Pro Video 34%
  14. 14 Cosmos3-Super-64B Video 33.1%
  15. 15 GLM-4.6V Video 33.1%
  16. 16 Gemma 4 26B A4B Video 32.8%
  17. 17 Qwen3-VL-4B Video 32.7%
  18. 18 GPT-5.6 Luna Video 32.5%
  19. 19 InternVideo3-8B Video 30.9%
  20. 20 InternVL3.5-30B-A3B Video 30.7%
  21. 21 RoboBrain2.5-8B Video 30.5%
  22. 22 InternVL3.5-8B Video 30.4%
  23. 23 GLM-4.6V-Flash Video 29%
  24. 24 SenseNova-SI-1.5-8B Video 28.8%
  25. 25 Video-XL-2-8B Video 28.7%
  26. 26 VST-7B-RL Video 28.4%
  27. 27 Flash-VStream-7B Video 27.1%
  28. 28 Cambrian-P-7B Video 26.5%
  29. 29 VideoChat-Online-4B Video 26.2%
  30. 30 Gemma 4 E4B Video 25%
  31. 31 Gemma 4 E2B Video 24.2%
  32. 32 Cambrian-S-7B Video 23.8%
  33. 33 Spatial-TTT-3B Video 22%

Source: The ReSpaceBench paper

#EntryScoreSettingDateSource
1 Qwen3-VL-8B-SFT 60.4% Video Official leaderboard
2 GPT-5.6 Sol 50.5% Video Jul 2026 Official leaderboard
3 Seed2.0 Pro 49.3% Video Feb 2026 Official leaderboard
4 Seed2.1 Pro 48% Video Jun 2026 Official leaderboard
5 Qwen 3.8 Max 46.3% Video Aug 2026 Official leaderboard
6 Qwen3-VL-32B 41.3% Video Official leaderboard
7 Qwen3.6 27B 40.3% Video Apr 2026 Official leaderboard
8 InternVL3.5-38B 38% Video Aug 2025 Official leaderboard
9 Qwen 3.6 35B-A3B 36.8% Video Apr 2026 Official leaderboard
10 Mage-VL 35.7% Video Official leaderboard
11 Qwen3-VL-8B-Instruct 35.5% Video Oct 2025 Official leaderboard
12 LLaVA-OneVision-2-8B 34.4% Video Official leaderboard
13 Gemini 3.1 Pro 34% Video Feb 2026 Official leaderboard
14 Cosmos3-Super-64B 33.1% Video Official leaderboard
15 GLM-4.6V 33.1% Video Dec 2025 Official leaderboard
16 Gemma 4 26B A4B 32.8% Video Apr 2026 Official leaderboard
17 Qwen3-VL-4B 32.7% Video Official leaderboard
18 GPT-5.6 Luna 32.5% Video Jul 2026 Official leaderboard
19 InternVideo3-8B 30.9% Video Official leaderboard
20 InternVL3.5-30B-A3B 30.7% Video Aug 2025 Official leaderboard
21 RoboBrain2.5-8B 30.5% Video Official leaderboard
22 InternVL3.5-8B 30.4% Video Aug 2025 Official leaderboard
23 GLM-4.6V-Flash 29% Video Dec 2025 Official leaderboard
24 SenseNova-SI-1.5-8B 28.8% Video Official leaderboard
25 Video-XL-2-8B 28.7% Video Official leaderboard
26 VST-7B-RL 28.4% Video Official leaderboard
27 Flash-VStream-7B 27.1% Video Official leaderboard
28 Cambrian-P-7B 26.5% Video Official leaderboard
29 VideoChat-Online-4B 26.2% Video Official leaderboard
30 Gemma 4 E4B 25% Video Mar 2026 Official leaderboard
31 Gemma 4 E2B 24.2% Video Mar 2026 Official leaderboard
32 Cambrian-S-7B 23.8% Video Official leaderboard
33 Spatial-TTT-3B 22% Video Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.6 Sol 50.5% Video Jul 2026 Official leaderboard
2 Seed2.0 Pro 49.3% Video Feb 2026 Official leaderboard
3 Seed2.1 Pro 48% Video Jun 2026 Official leaderboard
4 Qwen 3.8 Max 46.3% Video Aug 2026 Official leaderboard
5 Qwen3.6 27B 40.3% Video Apr 2026 Official leaderboard
6 InternVL3.5-38B 38% Video Aug 2025 Official leaderboard
7 Qwen 3.6 35B-A3B 36.8% Video Apr 2026 Official leaderboard
8 Qwen3-VL-8B-Instruct 35.5% Video Oct 2025 Official leaderboard
9 Gemini 3.1 Pro 34% Video Feb 2026 Official leaderboard
10 GLM-4.6V 33.1% Video Dec 2025 Official leaderboard
11 Gemma 4 26B A4B 32.8% Video Apr 2026 Official leaderboard
12 GPT-5.6 Luna 32.5% Video Jul 2026 Official leaderboard
13 InternVL3.5-30B-A3B 30.7% Video Aug 2025 Official leaderboard
14 InternVL3.5-8B 30.4% Video Aug 2025 Official leaderboard
15 GLM-4.6V-Flash 29% Video Dec 2025 Official leaderboard
16 Gemma 4 E4B 25% Video Mar 2026 Official leaderboard
17 Gemma 4 E2B 24.2% Video Mar 2026 Official leaderboard

Rankings by part

We read 12 parts from the benchmark’s paper, each ranking models on one part of the test, such as Query-to-memory association, Same-instance identification and Revisit counting. GPT-5.6 Sol leads 5 of them, and 4 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Query-to-memory association

17 models we track have a score here.

  1. 1 Seed2.1 Pro ↑2 58.8%
  2. 2 GPT-5.6 Sol ↓1 56.2%
  3. 3 Seed2.0 Pro ↓1 52.1%
  4. 4 Qwen 3.8 Max 50%
  5. 5 InternVL3.5-38B ↑1 44.2%
  6. 6 Gemini 3.1 Pro ↑3 44.2%
  7. 7 Qwen3-VL-8B-Instruct ↑1 43.1%
  8. 8 GLM-4.6V-Flash ↑7 42.3%
  9. 9 GLM-4.6V ↑1 38.8%
  10. 10 GPT-5.6 Luna ↑2 38.6%

Number one over time

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

OpenAI
Benchmark paper 1 leader
GPT-5.6 Sol 50.5%

Source: The ReSpaceBench 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% 2026 Near the ceiling (85%) GPT-5.6 Sol Seed2.0 Pro InternVL3.5-38B 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.6 Sol InternVL3.5-38B

Source: Official ReSpaceBench 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 Robotics

All 172 in Robotics
  • ERQA In use 400 multiple-choice visual questions from robot and egocentric footage on spatial, trajectory, action and state reasoning. Accuracy. Gemini Robotics ER 2 78.5%
  • SimplerEnv In use Real-to-sim replicas of Google Robot and WidowX/Bridge setups. Policies scored by success rate on pick, move, drawer and Bridge tasks. Praxis-1 80%
  • RoboLab In use Zero-shot DROID Franka policies in Isaac Lab on 120 tabletop tasks at three instruction levels, 10 episodes each. Success rate. Praxis-1 46.2%
  • RoboDojo In use Generalist manipulation policies run 42 Isaac Sim tasks and 18 remote real-robot tasks. Scored by task score and success rate. GPT-6 Astra 27.6%
  • Progress classification (Google DeepMind) In use Models assign each frame of a robot video to one of five task-progress bands from 0 to 100 percent. Scored by accuracy. Gemini Robotics ER 2 57.4%
  • Human proximity (Google DeepMind robotics) In use Models judge from robot camera views whether a human is within 1 meter, a spatial awareness check for safe robot behavior. Scored by accuracy. Gemini Robotics ER 2 93%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark

Catalog record 1 warning 1 pass
  • Warning Task access We don’t know whether the tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (33 entries).
  • Pass Models scored 17 models we track, plus 16 other entries on its leaderboard.
  • Pass Newest score The newest is from August 2026, for Qwen 3.8 Max.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.6 Sol (#14).
  • Pass Headroom The best is 50.5%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 18 rows for models we track and 16 other entries 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