# Spatial perception and reasoning on VSI-Bench (offline)

Source: [Spatial perception and reasoning on VSI-Bench (offline)](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-08T07:02:14.327Z

Evaluation of spatial perception and reasoning capabilities on the offline split of the VSI-Bench dataset, reporting performance across metrics including average score, object count accuracy, absolute/relative distance/direction errors, object/room size analysis, route planning, and approach order.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Spatial perception and reasoning |
| Dataset (including any stated split) | VSI-Bench (offline) |
| Metric | Average Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 15 of 15 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Average Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Average%20Score>): Higher is better; unit: SCALAR.

- [Object Count Accuracy](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Object%20Count%20Accuracy>): Higher is better; unit: PERCENT.

- [Absolute Distance Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Absolute%20Distance%20Score>): Higher is better; unit: SCALAR.

- [Object Size Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Object%20Size%20Score>): Higher is better; unit: SCALAR.

- [Room Size Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Room%20Size%20Score>): Higher is better; unit: SCALAR.

- [Relative Distance Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Relative%20Distance%20Score>): Higher is better; unit: SCALAR.

- [Relative Direction Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Relative%20Direction%20Score>): Higher is better; unit: SCALAR.

- [Route Planning Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Route%20Planning%20Score>): Higher is better; unit: SCALAR.

- [Approach Order Score](<https://www.sota2.com/research/sota/spatial-perception-and-reasoning-on-vsi-bench-offline.md?metric=Approach%20Order%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Average Score | Result date | Report paper |
| --- | --- | --- | --- |
| Stream3D-VLM-8B | 65.9 | Not recorded | 6a264c9790966ade919bea85 |
| VLM-3R-8B | 60.9 | Not recorded | 6a264c9790966ade919bea85 |
| Stream3D-VLM-4B | 55.2 | Not recorded | 6a264c9790966ade919bea85 |
| Gemini-2.5 Pro | 51.5 | Not recorded | 6a264c9790966ade919bea85 |
| VG LLM-8B | 50.7 | Not recorded | 6a264c9790966ade919bea85 |
| Spatial-MLLM-4B | 48.4 | Not recorded | 6a264c9790966ade919bea85 |
| SpaceR-7B | 45.5 | Not recorded | 6a264c9790966ade919bea85 |
| Gemini-1.5 Pro | 45.4 | Not recorded | 6a264c9790966ade919bea85 |
| LLaVA-NeXT-Video-72B | 40.9 | Not recorded | 6a264c9790966ade919bea85 |
| LLaVA-OneVision-72B | 40.2 | Not recorded | 6a264c9790966ade919bea85 |
| Qwen2.5-VL-72B | 37 | Not recorded | 6a264c9790966ade919bea85 |
| GPT-4o | 34 | Not recorded | 6a264c9790966ade919bea85 |
| Qwen2.5-VL-7B | 33 | Not recorded | 6a264c9790966ade919bea85 |
| VILA-1.5-40B | 31.2 | Not recorded | 6a264c9790966ade919bea85 |
| LongVILA-8B | 21.6 | Not recorded | 6a264c9790966ade919bea85 |

### Stream3D-VLM-8B

Average Score: 65.9. Result ID: 6a26681dfbc373592c22a0d0.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Implementation](<https://github.com/hanxunyu/Stream3D-VLM>)

Recorded evaluation configuration: {"Online":"true"}. This record does not assert independent reproduction.

### VLM-3R-8B

Average Score: 60.9. Result ID: 6a26681dfbc373592c22a0ce.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa790d1e209131cea7be1f.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### Stream3D-VLM-4B

Average Score: 55.2. Result ID: 6a26681dfbc373592c22a0cf.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Implementation](<https://github.com/hanxunyu/Stream3D-VLM>)

Recorded evaluation configuration: {"Online":"true"}. This record does not assert independent reproduction.

### Gemini-2.5 Pro

Average Score: 51.5. Result ID: 6a26681dfbc373592c22a0c4.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c311e209131cea8cef5.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### VG LLM-8B

Average Score: 50.7. Result ID: 6a26681dfbc373592c22a0cd.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### Spatial-MLLM-4B

Average Score: 48.4. Result ID: 6a26681dfbc373592c22a0cc.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78f11e209131cea7b58a.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### SpaceR-7B

Average Score: 45.5. Result ID: 6a26681dfbc373592c22a0cb.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a41e209131cea798a1.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### Gemini-1.5 Pro

Average Score: 45.4. Result ID: 6a26681dfbc373592c22a0c3.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### LLaVA-NeXT-Video-72B

Average Score: 40.9. Result ID: 6a26681dfbc373592c22a0ca.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### LLaVA-OneVision-72B

Average Score: 40.2. Result ID: 6a26681dfbc373592c22a0c9.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa756f1e209131cea67e58.md>)

[Implementation](<https://llava-vl.github.io/blog/llava-onevision>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### Qwen2.5-VL-72B

Average Score: 37. Result ID: 6a26681dfbc373592c22a0c8.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### GPT-4o

Average Score: 34. Result ID: 6a26681dfbc373592c22a0c2.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75681e209131cea67a98.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### Qwen2.5-VL-7B

Average Score: 33. Result ID: 6a26681dfbc373592c22a0c7.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d3c.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### VILA-1.5-40B

Average Score: 31.2. Result ID: 6a26681dfbc373592c22a0c6.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

### LongVILA-8B

Average Score: 21.6. Result ID: 6a26681dfbc373592c22a0c5.

[Reporting paper](<https://www.sota2.com/research/paper/6a264c9790966ade919bea85.md>)

Recorded evaluation configuration: {"Online":"false"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
