# LLM Agent Reasoning on WebArena

Source: [LLM Agent Reasoning on WebArena](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena>)

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

Data updated: 2026-03-10T11:22:02.386Z

Evaluation of LLM Agent Reasoning capabilities on the WebArena dataset, measuring success rate (Acc.), performance change (ΔAcc), step counts, and associated token usage.

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 | LLM Agent Reasoning |
| Dataset (including any stated split) | WebArena |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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

- [Accuracy](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [ΔAccuracy](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=%CE%94Accuracy>): Higher is better; unit: SCALAR.

- [Avg Steps](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=Avg%20Steps>): Lower is better; unit: SCALAR.

- [Total Tokens](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=Total%20Tokens>): Lower is better; unit: SIZE.

- [ΔTotal Tokens](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=%CE%94Total%20Tokens>): Lower is better; unit: SIZE.

- [Task Cost (Tokens)](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=Task%20Cost%20(Tokens)>): Lower is better; unit: SIZE.

- [ΔTask Cost (Tokens)](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=%CE%94Task%20Cost%20(Tokens)>): Lower is better; unit: SIZE.

- [Total Steps](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=Total%20Steps>): Lower is better; unit: SCALAR.

- [ΔStep Cost (Tokens)](<https://www.sota2.com/research/sota/llm-agent-reasoning-on-webarena.md?metric=%CE%94Step%20Cost%20(Tokens)>): Lower is better; unit: SIZE.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| ARES | 46.5 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Rule-Based (High) | 45 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Rule-Based (Medium) | 42.6 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Prompting-Based (Gemini 3 Pro) | 41.9 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Prompting-Based (GPT 5) | 41.1 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Rule-Based (Random) | 40.3 | Not recorded | 69afa5932d603e6de3e43ba8 |
| Rule-Based (Low) | 37.4 | Not recorded | 69afa5932d603e6de3e43ba8 |

### ARES

Accuracy: 46.5. Result ID: 69affeca6dbb7075b95967eb.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Adaptive Selection","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Rule-Based (High)

Accuracy: 45. Result ID: 69affeca6dbb7075b95967e7.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Rule-Based","Effort Level":"High","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Rule-Based (Medium)

Accuracy: 42.6. Result ID: 69affeca6dbb7075b95967e6.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Rule-Based","Effort Level":"Medium","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Prompting-Based (Gemini 3 Pro)

Accuracy: 41.9. Result ID: 69affeca6dbb7075b95967ea.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Prompting-Based","Model":"Gemini 3 Pro","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Prompting-Based (GPT 5)

Accuracy: 41.1. Result ID: 69affeca6dbb7075b95967e9.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

[Method origin paper](<https://www.sota2.com/research/paper/695f39f053fdd5c361743ecb.md>)

Recorded evaluation configuration: {"Strategy":"Prompting-Based","Model":"GPT 5","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Rule-Based (Random)

Accuracy: 40.3. Result ID: 69affeca6dbb7075b95967e8.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Rule-Based","Effort Level":"Random","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

### Rule-Based (Low)

Accuracy: 37.4. Result ID: 69affeca6dbb7075b95967e5.

[Reporting paper](<https://www.sota2.com/research/paper/69afa5932d603e6de3e43ba8.md>)

Recorded evaluation configuration: {"Strategy":"Rule-Based","Effort Level":"Low","Backbone LLM":"gpt-oss-20b"}. This record does not assert independent reproduction.

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