Skip to content

Published evaluation

WebVoyager

Read the actual model outputs. Follow each task from its instructions to the final response and evaluator verdict.

Model run report

Browser Use v4

gpt-5.6-luna · 2026-09-16

66.7%Task success10 / 15 tasks passed
15 / 643Tasks attemptedPublished task coverage
117.9sMean task time15 recorded measurements
$0.039Mean cost / task15 recorded measurements

Allrecipes

Allrecipes--1

Passed

Find a recipe for a vegetarian lasagna that has at least a four-star rating and uses zucchini.

Agent response

Debbie’s Vegetable Lasagna

  • Rating: 4.7/5 from 20 ratings
  • Vegetarian: Listed under Allrecipes’ “Vegetarian Lasagna Recipes”
  • Zucchini: Includes 2 cups coarsely chopped zucchini

Source: Allrecipes recipe

Original response text
**Debbie’s Vegetable Lasagna**

- **Rating:** 4.7/5 from 20 ratings
- **Vegetarian:** Listed under Allrecipes’ “Vegetarian Lasagna Recipes”
- **Zucchini:** Includes 2 cups coarsely chopped zucchini

Source: [Allrecipes recipe](https://www.allrecipes.com/recipe/236878/debbies-vegetable-lasagna/)
Run configuration & methodology
Run
webvoyager-browser-use-20260915-vision-20260916:browser-use-v4
Model
gpt-5.6-luna
Environment
Browser Use Cloud v4; fresh hosted browser per task; US proxy
Task budget
300s/task; $1.0 agent cap/task; one accepted run per task
Primary evaluator
rubric
Benchmark version
5a7896738c10bfb8b9edccce6bb0e0411f8ae569:69b19fd86c23f1a500244a3724e039aa7ca6a1223d03e11eb10e308d4f11c488
Evaluation
Primary: rubric; declared grading contract 82db14ea2dc5d3458d75a485c7de1eb4eefad97fca0871ea20146a4c8f0cb006

Imported recorded evidence; not independently rerun by the report engine. Scaffold: browser-use-cloud-v4 at SDK 3.11.3. Unknown measurements remain unknown. 15/643 tasks; one per website, seed 20260915. Separate Codex AI evidence review, not official GPT-4V grading or human adjudication; review model version was not retained. Unverified requirements receive no success credit under the recorded rubric. A later, separate Astra vision review runs on GKE using only the final screenshot and answer. Its restricted evidence scope is supplemental and does not replace the original rubric primary. Vision latency and subscription token usage are recorded separately; no dollar cost is inferred. Provider trajectory judgments are retained separately. Latency measures create-to-terminal agent time. Cost is final provider-reported run cost including asynchronous judge updates; browser hosting is separate. Original fixed dates were preserved.

Imported recorded evidence; not independently rerun by the report engine. Scaffold: browser-use-cloud-v4 at SDK 3.11.3. Unknown measurements remain unknown. 15/643 tasks; one per website, seed 20260915. Separate Codex AI evidence review, not official GPT-4V grading or human adjudication; review model version was not retained. Unverified requirements receive no success credit under the recorded rubric. A later, separate Astra vision review runs on GKE using only the final screenshot and answer. Its restricted evidence scope is supplemental and does not replace the original rubric primary. Vision latency and subscription token usage are recorded separately; no dollar cost is inferred. Provider trajectory judgments are retained separately. Latency measures create-to-terminal agent time. Cost is final provider-reported run cost including asynchronous judge updates; browser hosting is separate. Original fixed dates were preserved.

Environment

What Browser Use v4 had to work with.

Browser Use v4gpt-5.6-luna2026-09-16T00:31:33.909105+00:00 → 2026-09-16T00:43:45.396405+00:00

Agent & scaffold

Model
gpt-5.6-luna
Agent
Browser Use v4
Scaffold
browser-use-cloud-v4 · SDK 3.11.3
Harness
Browser Use Cloud · GKE vision review
Environment
Browser Use Cloud v4; fresh hosted browser per task; US proxy

Budget

Budget
300s/task; $1.0 agent cap/task; one accepted run per task
Recorded cost per task
$0.0392 · 15 of 15 attempted tasks

Grader

Primary evaluator
rubric
Evaluator
Rubric review
Verdicts by grader
codex-evidence-review-71e8d3f09ff1 · Supervising Codex assistant → 15 verdicts

Recorded window

Recorded window
2026-09-16T00:31:33.909105+00:00 → 2026-09-16T00:43:45.396405+00:00
Task roster
15 attempted of 643 tasks
Task source
5a7896738c10

Sample task from this run (1 of 15 attempted)

Allrecipes · Allrecipes--1

One real task from this run on WebVoyager: the prompt, the answer as given, and the grader’s verdict.

This sample is for illustration; a single task does not reflect a configuration’s overall result.

Task

Find a recipe for a vegetarian lasagna that has at least a four-star rating and uses zucchini.

Browser Use v4

gpt-5.6-lunaBrowser Use Cloud · GKE vision review41.5 s$0.0168

Pass

Graded by codex-evidence-review-71e8d3f09ff1 · Supervising Codex assistant · codex-evidence-review-71e8d3f09ff1

Grader reasoning

The captured recipe page identifies a vegetarian lasagna, shows a 4.7 rating, and lists zucchini among the ingredients. All three requested constraints are supported.

Trajectory

Browser Use v4

gpt-5.6-lunaBrowser Use Cloud · GKE vision review41.5 s$0.0168

Pass

Final answer

**Debbie’s Vegetable Lasagna**

- **Rating:** 4.7/5 from 20 ratings
- **Vegetarian:** Listed under Allrecipes’ “Vegetarian Lasagna Recipes”
- **Zucchini:** Includes 2 cups coarsely chopped zucchini

Source: [Allrecipes recipe](https://www.allrecipes.com/recipe/236878/debbies-vegetable-lasagna/)