# Agent Continuous Improvement on CoSQL StreamBench (32 held-out episodes)

Source: [Agent Continuous Improvement on CoSQL StreamBench (32 held-out episodes)](<https://www.sota2.com/research/sota/agent-continuous-improvement-on-cosql-streambench-32-held-out-episodes>)

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

Data updated: 2026-06-09T09:13:48.661Z

Evaluation of language agent performance focusing on continuous improvement capabilities, specifically tested using 32 held-out episodes from the CoSQL StreamBench dataset, measured by metrics like pass@k.

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 | Continuous improvement of language agents |
| Dataset (including any stated split) | CoSQL StreamBench (32 held-out episodes) |
| Metric | Pass@4 |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 5 of 5 |
| 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

- [Pass@4](<https://www.sota2.com/research/sota/agent-continuous-improvement-on-cosql-streambench-32-held-out-episodes.md?metric=Pass%404>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Pass@4 | Result date | Report paper |
| --- | --- | --- | --- |
| Memory w/ MEMOPILOT | 73.5 | Not recorded | 6a279e2890966ade919bf7e5 |
| Full History | 70 | Not recorded | 6a279e2890966ade919bf7e5 |
| No Memory | 69.5 | Not recorded | 6a279e2890966ade919bf7e5 |
| Memory w/ DeepSeek-V3.2 | 67.5 | Not recorded | 6a279e2890966ade919bf7e5 |
| Memory w/ Qwen2.5-14B | 66 | Not recorded | 6a279e2890966ade919bf7e5 |

### Memory w/ MEMOPILOT

Pass@4: 73.5. Result ID: 6a27d8dcfbc373592c24d565.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

Recorded evaluation configuration: {"Memory Strategy":"MEMOPILOT","Execution Agent":"Qwen2.5-14B-Instruct"}. This record does not assert independent reproduction.

### Full History

Pass@4: 70. Result ID: 6a27d8dcfbc373592c24d562.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

Recorded evaluation configuration: {"Memory Strategy":"Full History","Execution Agent":"Qwen2.5-14B-Instruct"}. This record does not assert independent reproduction.

### No Memory

Pass@4: 69.5. Result ID: 6a27d8dcfbc373592c24d561.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

Recorded evaluation configuration: {"Memory Strategy":"No Memory","Execution Agent":"Qwen2.5-14B-Instruct"}. This record does not assert independent reproduction.

### Memory w/ DeepSeek-V3.2

Pass@4: 67.5. Result ID: 6a27d8dcfbc373592c24d563.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

Recorded evaluation configuration: {"Memory Strategy":"DeepSeek-V3.2","Execution Agent":"Qwen2.5-14B-Instruct"}. This record does not assert independent reproduction.

### Memory w/ Qwen2.5-14B

Pass@4: 66. Result ID: 6a27d8dcfbc373592c24d564.

[Reporting paper](<https://www.sota2.com/research/paper/6a279e2890966ade919bf7e5.md>)

Recorded evaluation configuration: {"Memory Strategy":"Qwen2.5-14B","Execution Agent":"Qwen2.5-14B-Instruct"}. This record does not assert independent reproduction.

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