# Text Generation Quality and Watermark Detectability on ELI5

Source: [Text Generation Quality and Watermark Detectability on ELI5](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5>)

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

Data updated: 2026-02-26T05:25:07.635Z

Evaluation of text generation models on the ELI5 dataset, assessing overall quality (PPL, MAUVE) alongside watermark detection performance (AUC, TPR).

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 | Text generation quality and watermark detectability |
| Dataset (including any stated split) | ELI5 |
| Metric | AUC |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 16 of 16 |
| 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

- [AUC](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5.md?metric=AUC>): Higher is better; unit: PERCENT.

- [TPR](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5.md?metric=TPR>): Higher is better; unit: PERCENT.

- [PPL](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5.md?metric=PPL>): Lower is better; unit: SCALAR.

- [Seq-rep-3](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5.md?metric=Seq-rep-3>): Lower is better; unit: PERCENT.

- [MAUVE](<https://www.sota2.com/research/sota/text-generation-quality-and-watermark-detectability-on-eli5.md?metric=MAUVE>): Higher is better; unit: PERCENT.

## Recorded results

| Method | AUC | Result date | Report paper |
| --- | --- | --- | --- |
| KGW-2 | 100 | Not recorded | 693115d9d6b0dd3342933da6 |
| Gumbel-max | 99.8 | Not recorded | 693115d9d6b0dd3342933da6 |
| SynthID | 99.7 | Not recorded | 693115d9d6b0dd3342933da6 |
| KGW-2 | 99.6 | Not recorded | 693115d9d6b0dd3342933da6 |
| MarkTune | 99.6 | Not recorded | 693115d9d6b0dd3342933da6 |
| SynthID | 99.3 | Not recorded | 693115d9d6b0dd3342933da6 |
| KGW-1 | 99.3 | Not recorded | 693115d9d6b0dd3342933da6 |
| MarkTune | 99.3 | Not recorded | 693115d9d6b0dd3342933da6 |
| Gumbel-max | 99.1 | Not recorded | 693115d9d6b0dd3342933da6 |
| GaussMark | 97.9 | Not recorded | 693115d9d6b0dd3342933da6 |
| KGW-1 | 97.1 | Not recorded | 693115d9d6b0dd3342933da6 |
| KGW-D | 96.9 | Not recorded | 693115d9d6b0dd3342933da6 |
| GaussMark | 95.8 | Not recorded | 693115d9d6b0dd3342933da6 |
| KGW-D | 94.1 | Not recorded | 693115d9d6b0dd3342933da6 |
| No Watermark | 50 | Not recorded | 693115d9d6b0dd3342933da6 |
| No Watermark | 50 | Not recorded | 693115d9d6b0dd3342933da6 |

### KGW-2

AUC: 100. Result ID: 69998b265698665ec4fd0d81.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73131e209131cea5a4b7.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### Gumbel-max

AUC: 99.8. Result ID: 69998b265698665ec4fd0d82.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### SynthID

AUC: 99.7. Result ID: 69998b265698665ec4fd0d83.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### KGW-2

AUC: 99.6. Result ID: 69998b265698665ec4fd0d79.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73131e209131cea5a4b7.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### MarkTune

AUC: 99.6. Result ID: 69998b265698665ec4fd0d7e.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### SynthID

AUC: 99.3. Result ID: 69998b265698665ec4fd0d7b.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### KGW-1

AUC: 99.3. Result ID: 69998b265698665ec4fd0d80.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73131e209131cea5a4b7.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### MarkTune

AUC: 99.3. Result ID: 69998b265698665ec4fd0d86.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### Gumbel-max

AUC: 99.1. Result ID: 69998b265698665ec4fd0d7a.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### GaussMark

AUC: 97.9. Result ID: 69998b265698665ec4fd0d7d.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b31e209131cea6e8eb.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### KGW-1

AUC: 97.1. Result ID: 69998b265698665ec4fd0d78.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73131e209131cea5a4b7.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### KGW-D

AUC: 96.9. Result ID: 69998b265698665ec4fd0d84.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73141e209131cea5a619.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### GaussMark

AUC: 95.8. Result ID: 69998b265698665ec4fd0d85.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76b31e209131cea6e8eb.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

### KGW-D

AUC: 94.1. Result ID: 69998b265698665ec4fd0d7c.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa73141e209131cea5a619.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### No Watermark

AUC: 50. Result ID: 69998b265698665ec4fd0d77.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Qwen3-4B","Temperature (T)":0.7}. This record does not assert independent reproduction.

### No Watermark

AUC: 50. Result ID: 69998b265698665ec4fd0d7f.

[Reporting paper](<https://www.sota2.com/research/paper/693115d9d6b0dd3342933da6.md>)

Recorded evaluation configuration: {"Model":"Llama2-7B","Temperature (T)":1}. This record does not assert independent reproduction.

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