Compare runs

Pick two or more runs to compare. Gemma-4-31B on an unnamed computer is the first; add another below.

Runs of the same model, or on the same computer, the newest first.

The runs side by side: score, tasks passed, speed, tokens generated, each category and each task
RunGemma-4-31BNo computer named · Ollama v262Take out Gemma-4-31B
Score—No score
Tasks passed22 of 44
Points22 of 64
Speed13.1 tok/s
Tokens25,185
Summarizing5/5
One-sentence summaryPassed13.4 tok/s
Meeting notes in three bulletsPassed13.3 tok/s
Review sentiment and complaintPassed13.3 tok/s
Faithful summary of a studyPassed13.3 tok/s
Action items from an email threadPassed13.3 tok/s
Extracting4/5
Fields from an email as JSONPassed13.3 tok/s
Every email address, onceNot run
A list as CSVPassed13.3 tok/s
The real deadlinePassed13.3 tok/s
Warnings and errors from a logPassed13.2 tok/s
Following instructions3/5
Exactly one wordPassed13.4 tok/s
Three lowercase linesFailed13.3 tok/s
Exact JSON shapePassed13.3 tok/s
Sort a listPassed13.4 tok/s
A name with three rulesFailed12.9 tok/s
Reasoning0/5
Unit priceNot run
Days of the weekNot run
Who owns the catNot run
Arrival timeNot run
Filling a tankNot run
Reading code1/4
What does it printNot run
Trace a loopNot run
Find the bugPassed13.3 tok/s
Run a query in your headNot run
Using tools5/5
Call one toolPassed13.4 tok/s
Don't call a tool when none is neededPassed13.4 tok/s
Two calls for two answersPassed13.3 tok/s
Chain two toolsPassed13.3 tok/s
Handle a failed call honestlyPassed13.3 tok/s
Building pages4/5
A small websitePassed13.1 tok/s
A pricing table with a togglePassed13.1 tok/s
An accessible sign-up formPassed13.2 tok/s
A to-do app that remembersPassed13.1 tok/s
A bar chart in SVGNot run
Building games0/2
Build 2048Not run
Build SnakeNot run
Writing code0/3
Merge intervalsNot run
An LRU cacheNot run
Evaluate arithmeticNot run
Fixing code0/2
Fix the bookings moduleNot run
Fix the CSV readerNot run
Planning and data0/3
A query over hidden dataNot run
Schedule five meetingsNot run
A replacement in ten calls or fewerNot run

A check marks the best of each row. tok/s is tokens a second: how fast the model writes. How the score works.