# Multi-needle Retrieval on MRCR

Source: [Multi-needle Retrieval on MRCR](<https://www.sota2.com/research/sota/multi-needle-retrieval-on-mrcr>)

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

Data updated: 2026-02-26T05:50:18.035Z

Evaluation of model performance on the MRCR dataset for Multi-needle Retrieval tasks, measured by needle-specific success rates (2, 4, and 8 needles) and overall average.

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 | Multi-needle Retrieval |
| Dataset (including any stated split) | MRCR |
| Metric | MRCR 2-Needle Success |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| 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

- [MRCR 2-Needle Success](<https://www.sota2.com/research/sota/multi-needle-retrieval-on-mrcr.md?metric=MRCR%202-Needle%20Success>): Higher is better; unit: PERCENT.

- [MRCR 4-Needle Success](<https://www.sota2.com/research/sota/multi-needle-retrieval-on-mrcr.md?metric=MRCR%204-Needle%20Success>): Higher is better; unit: PERCENT.

- [MRCR 8-Needle Success](<https://www.sota2.com/research/sota/multi-needle-retrieval-on-mrcr.md?metric=MRCR%208-Needle%20Success>): Higher is better; unit: PERCENT.

- [MRCR Average Success Rate](<https://www.sota2.com/research/sota/multi-needle-retrieval-on-mrcr.md?metric=MRCR%20Average%20Success%20Rate>): Higher is better; unit: PERCENT.

## Recorded results

| Method | MRCR 2-Needle Success | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen2.5-32B + SPELL | 38 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-32B + RLVR | 36.7 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-14B + SPELL | 35 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-7B + SPELL | 34.5 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-32B | 27 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-14B | 23.1 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-7B + RLVR | 22 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-14B + RLVR | 20.9 | Not recorded | 68fa7bb41e209131cea8a096 |
| Qwen2.5-7B | 6.9 | Not recorded | 68fa7bb41e209131cea8a096 |

### Qwen2.5-32B + SPELL

MRCR 2-Needle Success: 38. Result ID: 699ae4cb5698665ec474eb68.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-32B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"SPELL"}. This record does not assert independent reproduction.

### Qwen2.5-32B + RLVR

MRCR 2-Needle Success: 36.7. Result ID: 699ae4cb5698665ec474eb67.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-32B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"RLVR"}. This record does not assert independent reproduction.

### Qwen2.5-14B + SPELL

MRCR 2-Needle Success: 35. Result ID: 699ae4cb5698665ec474eb65.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"SPELL"}. This record does not assert independent reproduction.

### Qwen2.5-7B + SPELL

MRCR 2-Needle Success: 34.5. Result ID: 699ae4cb5698665ec474eb62.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"SPELL"}. This record does not assert independent reproduction.

### Qwen2.5-32B

MRCR 2-Needle Success: 27. Result ID: 699ae4cb5698665ec474eb66.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-32B","Maximum input length":"16K","Maximum output length":"4K"}. This record does not assert independent reproduction.

### Qwen2.5-14B

MRCR 2-Needle Success: 23.1. Result ID: 699ae4cb5698665ec474eb63.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B","Maximum input length":"16K","Maximum output length":"4K"}. This record does not assert independent reproduction.

### Qwen2.5-7B + RLVR

MRCR 2-Needle Success: 22. Result ID: 699ae4cb5698665ec474eb61.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"RLVR"}. This record does not assert independent reproduction.

### Qwen2.5-14B + RLVR

MRCR 2-Needle Success: 20.9. Result ID: 699ae4cb5698665ec474eb64.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-14B","Maximum input length":"16K","Maximum output length":"4K","Alignment":"RLVR"}. This record does not assert independent reproduction.

### Qwen2.5-7B

MRCR 2-Needle Success: 6.9. Result ID: 699ae4cb5698665ec474eb60.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7bb41e209131cea8a096.md>)

Recorded evaluation configuration: {"Backbone":"Qwen2.5-7B","Maximum input length":"16K","Maximum output length":"4K"}. This record does not assert independent reproduction.

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