# Task Accuracy on OSWorld

Source: [Task Accuracy on OSWorld](<https://www.sota2.com/research/sota/task-accuracy-on-osworld>)

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

Data updated: 2026-05-19T09:33:53.651Z

Evaluation of agent performance based on Task Accuracy within the OSWorld environment, measured by the overall accuracy score.

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 | Task accuracy |
| Dataset (including any stated split) | OSWorld |
| Metric | Task Accuracy |
| Unit | PERCENT |
| 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

- [Task Accuracy](<https://www.sota2.com/research/sota/task-accuracy-on-osworld.md?metric=Task%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Task Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| KIVI | 41.49 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV INT4 | 41.43 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV Mixed | 40.59 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang FP4 | 40.54 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang BF16 | 39.2 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV Mixed | 35.78 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| KIVI | 34.61 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang BF16 | 34.02 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV INT4 | 33.92 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang FP4 | 29.52 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV INT4 | 13.11 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| TriAxialKV Mixed | 12.77 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| KIVI | 12.26 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang BF16 | 12.22 | Not recorded | 6a0bee91dbc836b87d82c3b3 |
| SGLang FP4 | 12.13 | Not recorded | 6a0bee91dbc836b87d82c3b3 |

### KIVI

Task Accuracy: 41.49. Result ID: 6a0c2e73fbc373592c029e0f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa744b1e209131cea60dab.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-32B-Thinking"}. This record does not assert independent reproduction.

### TriAxialKV INT4

Task Accuracy: 41.43. Result ID: 6a0c2e73fbc373592c029e12.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-32B-Thinking"}. This record does not assert independent reproduction.

### TriAxialKV Mixed

Task Accuracy: 40.59. Result ID: 6a0c2e73fbc373592c029e15.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-32B-Thinking"}. This record does not assert independent reproduction.

### SGLang FP4

Task Accuracy: 40.54. Result ID: 6a0c2e73fbc373592c029e0c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-32B-Thinking"}. This record does not assert independent reproduction.

### SGLang BF16

Task Accuracy: 39.2. Result ID: 6a0c2e73fbc373592c029e09.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-32B-Thinking"}. This record does not assert independent reproduction.

### TriAxialKV Mixed

Task Accuracy: 35.78. Result ID: 6a0c2e73fbc373592c029e14.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-8B-Thinking"}. This record does not assert independent reproduction.

### KIVI

Task Accuracy: 34.61. Result ID: 6a0c2e73fbc373592c029e0e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa744b1e209131cea60dab.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-8B-Thinking"}. This record does not assert independent reproduction.

### SGLang BF16

Task Accuracy: 34.02. Result ID: 6a0c2e73fbc373592c029e08.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-8B-Thinking"}. This record does not assert independent reproduction.

### TriAxialKV INT4

Task Accuracy: 33.92. Result ID: 6a0c2e73fbc373592c029e11.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-8B-Thinking"}. This record does not assert independent reproduction.

### SGLang FP4

Task Accuracy: 29.52. Result ID: 6a0c2e73fbc373592c029e0b.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL-8B-Thinking"}. This record does not assert independent reproduction.

### TriAxialKV INT4

Task Accuracy: 13.11. Result ID: 6a0c2e73fbc373592c029e13.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### TriAxialKV Mixed

Task Accuracy: 12.77. Result ID: 6a0c2e73fbc373592c029e16.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

Recorded evaluation configuration: {"Model":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### KIVI

Task Accuracy: 12.26. Result ID: 6a0c2e73fbc373592c029e10.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa744b1e209131cea60dab.md>)

Recorded evaluation configuration: {"Model":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### SGLang BF16

Task Accuracy: 12.22. Result ID: 6a0c2e73fbc373592c029e0a.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"InternVL3.5-38B"}. This record does not assert independent reproduction.

### SGLang FP4

Task Accuracy: 12.13. Result ID: 6a0c2e73fbc373592c029e0d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee91dbc836b87d82c3b3.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa738a1e209131cea5d174.md>)

Recorded evaluation configuration: {"Model":"InternVL3.5-38B"}. This record does not assert independent reproduction.

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