# Reading Comprehension on BELEBELE (Korean, Bengali, Swahili, Telugu)

Source: [Reading Comprehension on BELEBELE (Korean, Bengali, Swahili, Telugu)](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu>)

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

Data updated: 2026-02-26T05:35:34.560Z

Evaluation of Reading Comprehension performance across multiple languages (Korean, Bengali, Swahili, Telugu) within the BELEBELE benchmark suite, reporting language-specific and average scores.

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 | Reading Comprehension |
| Dataset (including any stated split) | BELEBELE |
| Metric | Average RC Score (BELEBELE) |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 31 of 31 |
| 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 RC Score (BELEBELE)](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu.md?metric=Average%20RC%20Score%20(BELEBELE)>): Higher is better; unit: PERCENT.

- [Korean RC Score](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu.md?metric=Korean%20RC%20Score>): Higher is better; unit: PERCENT.

- [Bengali RC Score](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu.md?metric=Bengali%20RC%20Score>): Higher is better; unit: PERCENT.

- [Swahili RC Score](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu.md?metric=Swahili%20RC%20Score>): Higher is better; unit: PERCENT.

- [Telugu RC Score](<https://www.sota2.com/research/sota/reading-comprehension-on-belebele-korean-bengali-swahili-telugu.md?metric=Telugu%20RC%20Score>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Average RC Score (BELEBELE) | Result date | Report paper |
| --- | --- | --- | --- |
| Trinity Large (MoE) | 80 | Not recorded | 6995476c09c33900fcd6d4aa |
| Qwen3 8B | 77 | Not recorded | 6995476c09c33900fcd6d4aa |
| Qwen3 4B | 71 | Not recorded | 6995476c09c33900fcd6d4aa |
| Module-wise Swapping | 70.6 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Layer-wise Swapping | 70.5 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Task Arithmetic | 70.4 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Ties Merging | 70.4 | Not recorded | 69802f65a8b01e76cc8e14bf |
| DARE | 70.4 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Safety-Only | 70.3 | Not recorded | 69802f65a8b01e76cc8e14bf |
| base | 70.2 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Lang-Only | 70.1 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Layer Swapping | 70.1 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Mixed SFT | 70 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Lang-Only | 56.3 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Layer Swapping | 56.3 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Layer-wise Swapping | 56.2 | Not recorded | 69802f65a8b01e76cc8e14bf |
| DARE | 56.1 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Module-wise Swapping | 55.8 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Task Arithmetic | 55.7 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Ties Merging | 55.6 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Mixed SFT | 55.5 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Sarvam1 2B | 55 | Not recorded | 6995476c09c33900fcd6d4aa |
| Safety-Only | 54.2 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Llama-3.1 8B | 53 | Not recorded | 6995476c09c33900fcd6d4aa |
| Datology 8B | 51 | Not recorded | 6995476c09c33900fcd6d4aa |
| Granite-4.0 Micro | 51 | Not recorded | 6995476c09c33900fcd6d4aa |
| base | 50.1 | Not recorded | 69802f65a8b01e76cc8e14bf |
| Datology 3B | 43 | Not recorded | 6995476c09c33900fcd6d4aa |
| Llama-3.2 3B | 39 | Not recorded | 6995476c09c33900fcd6d4aa |
| SmolLM3 3B | 30 | Not recorded | 6995476c09c33900fcd6d4aa |
| Llama-3.2 1B | 25 | Not recorded | 6995476c09c33900fcd6d4aa |

### Trinity Large (MoE)

Average RC Score (BELEBELE): 80. Result ID: 699ab0c55698665ec43e3405.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Architecture":"MoE","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Qwen3 8B

Average RC Score (BELEBELE): 77. Result ID: 699ab0c55698665ec43e3402.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"8B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Qwen3 4B

Average RC Score (BELEBELE): 71. Result ID: 699ab0c55698665ec43e3401.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa788e1e209131cea79116.md>)

Recorded evaluation configuration: {"Parameters":"4B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Module-wise Swapping

Average RC Score (BELEBELE): 70.6. Result ID: 69965d4c5698665ec4a30531.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Layer-wise Swapping

Average RC Score (BELEBELE): 70.5. Result ID: 69965d4c5698665ec4a30530.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Task Arithmetic

Average RC Score (BELEBELE): 70.4. Result ID: 69965d4c5698665ec4a3052c.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ef01e209131cea43012.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Ties Merging

Average RC Score (BELEBELE): 70.4. Result ID: 69965d4c5698665ec4a3052d.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71091e209131cea4ecd8.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### DARE

Average RC Score (BELEBELE): 70.4. Result ID: 69965d4c5698665ec4a3052e.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73921e209131cea5d4aa.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Safety-Only

Average RC Score (BELEBELE): 70.3. Result ID: 69965d4c5698665ec4a3052a.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### base

Average RC Score (BELEBELE): 70.2. Result ID: 69965d4c5698665ec4a30528.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Lang-Only

Average RC Score (BELEBELE): 70.1. Result ID: 69965d4c5698665ec4a30529.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Layer Swapping

Average RC Score (BELEBELE): 70.1. Result ID: 69965d4c5698665ec4a3052f.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Mixed SFT

Average RC Score (BELEBELE): 70. Result ID: 69965d4c5698665ec4a3052b.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"Qwen 2 7B","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Lang-Only

Average RC Score (BELEBELE): 56.3. Result ID: 69965d4c5698665ec4a3051f.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Layer Swapping

Average RC Score (BELEBELE): 56.3. Result ID: 69965d4c5698665ec4a30525.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Layer-wise Swapping

Average RC Score (BELEBELE): 56.2. Result ID: 69965d4c5698665ec4a30526.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### DARE

Average RC Score (BELEBELE): 56.1. Result ID: 69965d4c5698665ec4a30524.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73921e209131cea5d4aa.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Module-wise Swapping

Average RC Score (BELEBELE): 55.8. Result ID: 69965d4c5698665ec4a30527.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Task Arithmetic

Average RC Score (BELEBELE): 55.7. Result ID: 69965d4c5698665ec4a30522.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ef01e209131cea43012.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Ties Merging

Average RC Score (BELEBELE): 55.6. Result ID: 69965d4c5698665ec4a30523.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71091e209131cea4ecd8.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Mixed SFT

Average RC Score (BELEBELE): 55.5. Result ID: 69965d4c5698665ec4a30521.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Sarvam1 2B

Average RC Score (BELEBELE): 55. Result ID: 699ab0c55698665ec43e3406.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"2B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Safety-Only

Average RC Score (BELEBELE): 54.2. Result ID: 69965d4c5698665ec4a30520.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Llama-3.1 8B

Average RC Score (BELEBELE): 53. Result ID: 699ab0c55698665ec43e3400.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"8B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Datology 8B

Average RC Score (BELEBELE): 51. Result ID: 699ab0c55698665ec43e33fd.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"8B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Granite-4.0 Micro

Average RC Score (BELEBELE): 51. Result ID: 699ab0c55698665ec43e3404.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"Micro","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### base

Average RC Score (BELEBELE): 50.1. Result ID: 69965d4c5698665ec4a3051e.

[Reporting paper](<https://www.sota2.com/research/paper/69802f65a8b01e76cc8e14bf.md>)

Recorded evaluation configuration: {"Backbone":"LLaMA 3.1 8B it","Evaluation Protocol":"0-shot"}. This record does not assert independent reproduction.

### Datology 3B

Average RC Score (BELEBELE): 43. Result ID: 699ab0c55698665ec43e33fc.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"3B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Llama-3.2 3B

Average RC Score (BELEBELE): 39. Result ID: 699ab0c55698665ec43e33ff.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"3B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### SmolLM3 3B

Average RC Score (BELEBELE): 30. Result ID: 699ab0c55698665ec43e3403.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"3B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

### Llama-3.2 1B

Average RC Score (BELEBELE): 25. Result ID: 699ab0c55698665ec43e33fe.

[Reporting paper](<https://www.sota2.com/research/paper/6995476c09c33900fcd6d4aa.md>)

Recorded evaluation configuration: {"Parameters":"1B","Protocol":"Zero-shot"}. This record does not assert independent reproduction.

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