Better at the work.
Built on evidence.

1. 

   AsidePublished · 300 tasks
   99.0%
2. 

   Browser UseReference benchmark panel
   97.7%
3. 

   Cook BrowserDesktop · adjusted cohort
   93.4%
4. 

   GPT-5.4Reference benchmark panel
   92.8%
5. 

   Claude Opus 4.8Reference benchmark panel
   84.0%
6. 

   ChatGPT AtlasReference benchmark panel
   70.0%

Cook: 268/287 after 13 impossible tasks were excluded; 268/300 (89.3%) across the full cohort. Merged baseline and targeted reruns. Aside: 297/300. Different runs and scoring policies; published reference, not a controlled head-to-head. Other competitors reproduce the reference benchmark panel; evaluation setups differ.

1. 

   Cook BrowserDesktop · 100 tasks · best run per task
   94.0%
2. 

   AsidePublished · 100 tasks
   93.0%
3. 

   Browser UseReference benchmark panel
   89.5%
4. 

   Claude Fable 5Reference benchmark panel
   80.0%
5. 

   GPT-5.5Reference benchmark panel
   68.0%
6. 

   Gemini 3.6 FlashReference benchmark panel
   58.0%

Cook: 94/100 tasks passed in their best graded run (up to 9 graded runs per task, 1.64 on average); 84/100 (84.0%) in each task's first graded run. Not counted: 1 repeat run that an LLM judge found reused an earlier attempt's saved work. September 2026, full suite, official BU Bench judge from Aside's repository. Aside: 93/100 published. Different actors and run policies; published reference, not a controlled head-to-head. Other competitors reproduce the reference benchmark panel; evaluation setups differ.

1. 

   Cook BrowserDesktop · 200 tasks · best run per task
   95.6%
2. 

   AsidePublished · rubric items
   88.8%
3. 

   Browser UseReference benchmark panel
   70.0%
4. 

   WebWrightReference benchmark panel
   60.8%
5. 

   Claude Opus 4.6Reference benchmark panel
   44.5%
6. 

   GPT-5.4Reference benchmark panel
   33.5%

Cook: 1,171/1,225 rubric items in each task's best graded run across all 200 tasks (up to 29 graded runs per task); 867/1,225 (70.8%) in the first graded run; 182/200 tasks met every item (4 of them in runs where Cook's desktop agent drove Google Chrome on a Mac instead of Cook's own browser engine). Not counted: 28 repeat runs that an LLM judge found reused an earlier attempt's saved work. September 2026, official Odysseys judge from Aside's repository. Aside: 1,050/1,182 rubric items across 200 tasks. Rubric totals and run policies differ; context, not an equivalent evaluation. Other competitors reproduce the reference benchmark panel; evaluation setups differ.

1. 

   Cook MemoryAnswer-bearing excerpts
   98.5%
2. m

   agentmemoryPublished · gold sessions
   95.2%

Cook: 476 answerable questions, scored on answer-bearing excerpts. agentmemory: published 500-question session-retrieval evaluation. Different retrieval units and cohorts; reference comparison, not an identical-harness rerun. Retrieval is separate from final-answer accuracy.

1. 

   Cook MemoryAnswer-bearing excerpts
   98.9%
2. m

   agentmemoryPublished · gold sessions
   98.6%

Cook: 476 answerable questions. agentmemory: published 500-question evaluation. Cook retrieves exchanges; agentmemory retrieves sessions. Scores measure finding relevant evidence, not whether the final answer is correct.

1. 

   Cook MemoryAnswer-bearing excerpts
   90.2
2. m

   agentmemoryPublished · gold sessions
   88.2

MRR rewards placing the first relevant result earlier: rank 1 contributes 1, rank 2 contributes 1/2. Shown ×100. Cook uses answer-bearing excerpts; agentmemory uses gold sessions. Published results from separate evaluation setups.