ReSpaceBench
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.
- Made by
- Tsinghua University and ByteDance
- Released
- November 2026
- Kind
- Embodied reasoning
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Multiple choice
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Generated, Real work and use
- 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
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:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 15 more
- 60.4 Qwen3-VL-8B-SFT
- 50.5 GPT-5.6 Sol
- 49.3
Seed2.0 Pro
- 48.0
Seed2.1 Pro
- 46.3
Qwen 3.8 Max
- 41.3 Qwen3-VL-32B
- 40.3
Qwen3.6 27B
- 38.0
InternVL3.5-38B
- 36.8
Qwen 3.6 35B-A3B
- 35.7 Mage-VL
- 35.5
Qwen3-VL-8B-Instruct
- 34.4 LLaVA-OneVision-2-8B
- 34.0
Gemini 3.1 Pro
- 33.1 Cosmos3-Super-64B
- 33.1 GLM-4.6V
- 32.8
Gemma 4 26B A4B
- 32.7 Qwen3-VL-4B
- 32.5 GPT-5.6 Luna
- 30.9 InternVideo3-8B
- 30.7
InternVL3.5-30B-A3B
- 30.5 RoboBrain2.5-8B
- 30.4
InternVL3.5-8B
- 29.0 GLM-4.6V-Flash
- 28.8 SenseNova-SI-1.5-8B
- 28.7 Video-XL-2-8B
- 28.4 VST-7B-RL
- 27.1 Flash-VStream-7B
- 26.5 Cambrian-P-7B
- 26.2 VideoChat-Online-4B
- 25.0
Gemma 4 E4B
- 24.2
Gemma 4 E2B
- 23.8 Cambrian-S-7B
- 22.0 Spatial-TTT-3B
- 1 Qwen3-VL-8B-SFT Video 60.4%
- 2 GPT-5.6 Sol Video 50.5%
- 3
Seed2.0 Pro Video 49.3%
- 4
Seed2.1 Pro Video 48%
- 5
Qwen 3.8 Max Video 46.3%
- 6 Qwen3-VL-32B Video 41.3%
- 7
Qwen3.6 27B Video 40.3%
- 8
InternVL3.5-38B Video 38%
- 9
Qwen 3.6 35B-A3B Video 36.8%
- 10 Mage-VL Video 35.7%
- 11
Qwen3-VL-8B-Instruct Video 35.5%
- 12 LLaVA-OneVision-2-8B Video 34.4%
- 13
Gemini 3.1 Pro Video 34%
- 14 Cosmos3-Super-64B Video 33.1%
- 15 GLM-4.6V Video 33.1%
- 16
Gemma 4 26B A4B Video 32.8%
- 17 Qwen3-VL-4B Video 32.7%
- 18 GPT-5.6 Luna Video 32.5%
- 19 InternVideo3-8B Video 30.9%
- 20
InternVL3.5-30B-A3B Video 30.7%
- 21 RoboBrain2.5-8B Video 30.5%
- 22
InternVL3.5-8B Video 30.4%
- 23 GLM-4.6V-Flash Video 29%
- 24 SenseNova-SI-1.5-8B Video 28.8%
- 25 Video-XL-2-8B Video 28.7%
- 26 VST-7B-RL Video 28.4%
- 27 Flash-VStream-7B Video 27.1%
- 28 Cambrian-P-7B Video 26.5%
- 29 VideoChat-Online-4B Video 26.2%
- 30
Gemma 4 E4B Video 25%
- 31
Gemma 4 E2B Video 24.2%
- 32 Cambrian-S-7B Video 23.8%
- 33 Spatial-TTT-3B Video 22%
Source: The ReSpaceBench paper
- 50.5 GPT-5.6 Sol
- 49.3
Seed2.0 Pro
- 48.0
Seed2.1 Pro
- 46.3
Qwen 3.8 Max
- 40.3
Qwen3.6 27B
- 38.0
InternVL3.5-38B
- 36.8
Qwen 3.6 35B-A3B
- 35.5
Qwen3-VL-8B-Instruct
- 34.0
Gemini 3.1 Pro
- 33.1 GLM-4.6V
- 32.8
Gemma 4 26B A4B
- 32.5 GPT-5.6 Luna
- 30.7
InternVL3.5-30B-A3B
- 30.4
InternVL3.5-8B
- 29.0 GLM-4.6V-Flash
- 25.0
Gemma 4 E4B
- 24.2
Gemma 4 E2B
- 1 GPT-5.6 Sol Video 50.5%
- 2
Seed2.0 Pro Video 49.3%
- 3
Seed2.1 Pro Video 48%
- 4
Qwen 3.8 Max Video 46.3%
- 5
Qwen3.6 27B Video 40.3%
- 6
InternVL3.5-38B Video 38%
- 7
Qwen 3.6 35B-A3B Video 36.8%
- 8
Qwen3-VL-8B-Instruct Video 35.5%
- 9
Gemini 3.1 Pro Video 34%
- 10 GLM-4.6V Video 33.1%
- 11
Gemma 4 26B A4B Video 32.8%
- 12 GPT-5.6 Luna Video 32.5%
- 13
InternVL3.5-30B-A3B Video 30.7%
- 14
InternVL3.5-8B Video 30.4%
- 15 GLM-4.6V-Flash Video 29%
- 16
Gemma 4 E4B Video 25%
- 17
Gemma 4 E2B Video 24.2%
Source: The ReSpaceBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Qwen3-VL-8B-SFT | 60.4% | Video | Official leaderboard | |
| 2 | GPT-5.6 Sol | 50.5% | Video | Jul 2026 | Official leaderboard |
| 3 | | 49.3% | Video | Feb 2026 | Official leaderboard |
| 4 | | 48% | Video | Jun 2026 | Official leaderboard |
| 5 | | 46.3% | Video | Aug 2026 | Official leaderboard |
| 6 | Qwen3-VL-32B | 41.3% | Video | Official leaderboard | |
| 7 | | 40.3% | Video | Apr 2026 | Official leaderboard |
| 8 | | 38% | Video | Aug 2025 | Official leaderboard |
| 9 | | 36.8% | Video | Apr 2026 | Official leaderboard |
| 10 | Mage-VL | 35.7% | Video | Official leaderboard | |
| 11 | | 35.5% | Video | Oct 2025 | Official leaderboard |
| 12 | LLaVA-OneVision-2-8B | 34.4% | Video | Official leaderboard | |
| 13 | | 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 | | 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 | | 30.7% | Video | Aug 2025 | Official leaderboard |
| 21 | RoboBrain2.5-8B | 30.5% | Video | Official leaderboard | |
| 22 | | 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 | | 25% | Video | Mar 2026 | Official leaderboard |
| 31 | | 24.2% | Video | Mar 2026 | Official leaderboard |
| 32 | Cambrian-S-7B | 23.8% | Video | Official leaderboard | |
| 33 | Spatial-TTT-3B | 22% | Video | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol | 50.5% | Video | Jul 2026 | Official leaderboard |
| 2 | | 49.3% | Video | Feb 2026 | Official leaderboard |
| 3 | | 48% | Video | Jun 2026 | Official leaderboard |
| 4 | | 46.3% | Video | Aug 2026 | Official leaderboard |
| 5 | | 40.3% | Video | Apr 2026 | Official leaderboard |
| 6 | | 38% | Video | Aug 2025 | Official leaderboard |
| 7 | | 36.8% | Video | Apr 2026 | Official leaderboard |
| 8 | | 35.5% | Video | Oct 2025 | Official leaderboard |
| 9 | | 34% | Video | Feb 2026 | Official leaderboard |
| 10 | GLM-4.6V | 33.1% | Video | Dec 2025 | Official leaderboard |
| 11 | | 32.8% | Video | Apr 2026 | Official leaderboard |
| 12 | GPT-5.6 Luna | 32.5% | Video | Jul 2026 | Official leaderboard |
| 13 | | 30.7% | Video | Aug 2025 | Official leaderboard |
| 14 | | 30.4% | Video | Aug 2025 | Official leaderboard |
| 15 | GLM-4.6V-Flash | 29% | Video | Dec 2025 | Official leaderboard |
| 16 | | 25% | Video | Mar 2026 | Official leaderboard |
| 17 | | 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
Seed2.1 Pro ↑2 58.8%
- 2 GPT-5.6 Sol ↓1 56.2%
- 3
Seed2.0 Pro ↓1 52.1%
- 4
Qwen 3.8 Max 50%
- 5
InternVL3.5-38B ↑1 44.2%
- 6
Gemini 3.1 Pro ↑3 44.2%
- 7
Qwen3-VL-8B-Instruct ↑1 43.1%
- 8 GLM-4.6V-Flash ↑7 42.3%
- 9 GLM-4.6V ↑1 38.8%
- 10 GPT-5.6 Luna ↑2 38.6%
Same-instance identification
17 models we track have a score here.
- 1 GPT-5.6 Sol 65.4%
- 2
Seed2.0 Pro 60.3%
- 3
Seed2.1 Pro 59%
- 4
Qwen 3.8 Max 49.4%
- 5
Qwen 3.6 35B-A3B ↑2 48.7%
- 6 GPT-5.6 Luna ↑6 44.2%
- 7
InternVL3.5-38B ↓1 40.4%
- 8
Qwen3.6 27B ↓3 40.4%
- 9
InternVL3.5-30B-A3B ↑4 39.7%
- 10
Gemini 3.1 Pro ↓1 39.7%
Revisit counting
17 models we track have a score here.
- 1 GPT-5.6 Sol 34.2%
- 2
Seed2.0 Pro 29.1%
- 3
Qwen3.6 27B ↑2 27.4%
- 4
Seed2.1 Pro ↓1 26.7%
- 5
Gemma 4 E4B ↑11 26.5%
- 6
Gemma 4 E2B ↑11 26.4%
- 7
InternVL3.5-8B ↑7 26.1%
- 8 GLM-4.6V ↑2 26%
- 9
Gemma 4 26B A4B ↑2 25.7%
- 10
InternVL3.5-38B ↓4 25.5%
Association
17 models we track have a score here.
- 1 GPT-5.6 Sol 51.9%
- 2
Seed2.1 Pro ↑1 48.2%
- 3
Seed2.0 Pro ↓1 47.2%
- 4
Qwen 3.8 Max 39.4%
- 5
InternVL3.5-38B ↑1 36.7%
- 6
Gemini 3.1 Pro ↑3 34.5%
- 7
Qwen3-VL-8B-Instruct ↑1 34.4%
- 8
Qwen 3.6 35B-A3B ↓1 34.4%
- 9
Qwen3.6 27B ↓4 34.1%
- 10 GLM-4.6V 34%
Room inventory
17 models we track have a score here.
- 1 GPT-5.6 Sol 55.4%
- 2
Seed2.0 Pro 48.8%
- 3
Qwen 3.8 Max ↑1 48.5%
- 4
Seed2.1 Pro ↓1 48.3%
- 5
Qwen3.6 27B 42.6%
- 6
Qwen3-VL-8B-Instruct ↑2 38.4%
- 7
Qwen 3.6 35B-A3B 38.2%
- 8
InternVL3.5-30B-A3B ↑5 35.5%
- 9 GLM-4.6V ↑1 32.8%
- 10
Gemma 4 26B A4B ↑1 32.3%
Object-place binding
17 models we track have a score here.
- 1 GPT-5.6 Sol 58.4%
- 2
Qwen 3.8 Max ↑2 56%
- 3
Qwen3.6 27B ↑2 55.4%
- 4
Seed2.1 Pro ↓1 54.4%
- 5
Seed2.0 Pro ↓3 52.3%
- 6
Qwen3-VL-8B-Instruct ↑2 42.7%
- 7 GPT-5.6 Luna ↑5 39%
- 8
Gemini 3.1 Pro ↑1 36.9%
- 9
Qwen 3.6 35B-A3B ↓2 35.3%
- 10
Gemma 4 26B A4B ↑1 34.5%
Unique-object counting
17 models we track have a score here.
- 1
InternVL3.5-38B ↑5 34.1%
- 2
Qwen3.6 27B ↑3 33.2%
- 3 GLM-4.6V ↑7 32.7%
- 4 GPT-5.6 Luna ↑8 32.7%
- 5
Seed2.0 Pro ↓3 31.8%
- 6
Seed2.1 Pro ↓3 30%
- 7 GPT-5.6 Sol ↓6 30%
- 8
InternVL3.5-8B ↑6 29.6%
- 9
InternVL3.5-30B-A3B ↑4 29.1%
- 10
Qwen 3.8 Max ↓6 27.4%
Cross-session state
17 models we track have a score here.
- 1
Qwen 3.8 Max ↑3 61.7%
- 2
Seed2.0 Pro 49.7%
- 3
Qwen3.6 27B ↑2 48.8%
- 4 GLM-4.6V ↑6 47.6%
- 5 GPT-5.6 Sol ↓4 47%
- 6
Seed2.1 Pro ↓3 46.3%
- 7
InternVL3.5-38B ↓1 44.5%
- 8
Gemini 3.1 Pro ↑1 44%
- 9
Qwen3-VL-8B-Instruct ↓1 41.5%
- 10
Qwen 3.6 35B-A3B ↓3 40.8%
Persistent state
17 models we track have a score here.
- 1
Qwen 3.8 Max ↑3 48.4%
- 2 GPT-5.6 Sol ↓1 47.7%
- 3
Seed2.0 Pro ↓1 45.6%
- 4
Qwen3.6 27B ↑1 45%
- 5
Seed2.1 Pro ↓2 44.8%
- 6
Qwen3-VL-8B-Instruct ↑2 37.4%
- 7
InternVL3.5-38B ↓1 35.4%
- 8
Qwen 3.6 35B-A3B ↓1 35%
- 9 GPT-5.6 Luna ↑3 34.6%
- 10
Gemma 4 26B A4B ↑1 33.1%
Spatial topology
17 models we track have a score here.
- 1
Seed2.1 Pro ↑2 44.2%
- 2
Seed2.0 Pro 43%
- 3
Qwen 3.8 Max ↑1 43%
- 4 GPT-5.6 Sol ↓3 41.9%
- 5
InternVL3.5-38B ↑1 34.9%
- 6
Qwen 3.6 35B-A3B ↑1 32.6%
- 7
Gemini 3.1 Pro ↑2 30.8%
- 8
Gemma 4 26B A4B ↑3 28.5%
- 9
InternVL3.5-8B ↑5 27.3%
- 10
Qwen3.6 27B ↓5 27.3%
Temporal successor
17 models we track have a score here.
- 1
Seed2.0 Pro ↑1 76.3%
- 2 GPT-5.6 Sol ↓1 66.2%
- 3
Seed2.1 Pro 64.7%
- 4
Qwen 3.8 Max 62.3%
- 5
InternVL3.5-38B ↑1 55.6%
- 6
Qwen 3.6 35B-A3B ↑1 55.6%
- 7
Qwen3.6 27B ↓2 53.1%
- 8
Gemini 3.1 Pro ↑1 45.4%
- 9 GLM-4.6V ↑1 44.4%
- 10
Gemma 4 26B A4B ↑1 44%
Reasoning
17 models we track have a score here.
- 1
Seed2.0 Pro ↑1 59.7%
- 2
Seed2.1 Pro ↑1 54.5%
- 3 GPT-5.6 Sol ↓2 54%
- 4
Qwen 3.8 Max 52.7%
- 5
InternVL3.5-38B ↑1 45.2%
- 6
Qwen 3.6 35B-A3B ↑1 44.1%
- 7
Qwen3.6 27B ↓2 40.2%
- 8
Gemini 3.1 Pro ↑1 38.1%
- 9
Gemma 4 26B A4B ↑2 36.2%
- 10 GLM-4.6V 35.3%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
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.
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
- 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.