# Web Search Agent on GAIA Avg@4

Source: [Web Search Agent on GAIA Avg@4](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4>)

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

Data updated: 2026-02-26T05:42:51.059Z

Evaluation of Web Search Agent capabilities on the specific GAIA Avg@4 configuration, reporting F1 and Last Jaccard (LasJ) scores across difficulty levels 1, 2, and 3.

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 | Web Search Agent |
| Dataset (including any stated split) | GAIA Avg@4 |
| Metric | F1 (Lv.1) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 4 of 4 |
| 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

- [F1 (Lv.1)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=F1%20(Lv.1)>): Higher is better; unit: PERCENT.

- [LasJ (Lv.1)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=LasJ%20(Lv.1)>): Higher is better; unit: PERCENT.

- [F1 (Lv.2)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=F1%20(Lv.2)>): Higher is better; unit: PERCENT.

- [LasJ (Lv.2)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=LasJ%20(Lv.2)>): Higher is better; unit: PERCENT.

- [F1 (Lv.3)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=F1%20(Lv.3)>): Higher is better; unit: PERCENT.

- [LasJ (Lv.3)](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=LasJ%20(Lv.3)>): Higher is better; unit: PERCENT.

- [Avg F1](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=Avg%20F1>): Higher is better; unit: PERCENT.

- [Avg LasJ](<https://www.sota2.com/research/sota/web-search-agent-on-gaia-avg-at-4.md?metric=Avg%20LasJ>): Higher is better; unit: PERCENT.

## Recorded results

| Method | F1 (Lv.1) | Result date | Report paper |
| --- | --- | --- | --- |
| BranPO | 57.9 | Not recorded | 6982d2f6a8b01e76cc8e99ab |
| GRPO | 53.5 | Not recorded | 6982d2f6a8b01e76cc8e99ab |
| BranPO | 41.2 | Not recorded | 6982d2f6a8b01e76cc8e99ab |
| GRPO | 38.2 | Not recorded | 6982d2f6a8b01e76cc8e99ab |

### BranPO

F1 (Lv.1): 57.9. Result ID: 69964eba5698665ec4935eeb.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

Recorded evaluation configuration: {"Base Model":"Qwen3-4B-Instruct-2507"}. This record does not assert independent reproduction.

### GRPO

F1 (Lv.1): 53.5. Result ID: 69964eba5698665ec4935eea.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Base Model":"Qwen3-4B-Instruct-2507"}. This record does not assert independent reproduction.

### BranPO

F1 (Lv.1): 41.2. Result ID: 69964eba5698665ec4935ee9.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction.

### GRPO

F1 (Lv.1): 38.2. Result ID: 69964eba5698665ec4935ee8.

[Reporting paper](<https://www.sota2.com/research/paper/6982d2f6a8b01e76cc8e99ab.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73101e209131cea5a395.md>)

[Implementation](<https://github.com/deepseek-ai/DeepSeek-Math>)

Recorded evaluation configuration: {"Base Model":"Qwen2.5-7B-Instruct"}. This record does not assert independent reproduction.

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