Agent BenchWe give models the same task and publish what it cost them.

Model

gemini-3.5-flash

5 of 5 tasks passed, in 12,842 ms of wall time and 2,260 thinking tokens.

Tasks passed
5 of 5
Total time
12,842 ms
Thinking tokens
2,260
One-word probe
1,288 ms
Not asked
4 (quota)
TaskResultTime Thinking tokensWhat came back
Output the number and nothing elsepass1,499ms155clean
Find one fact buried in a long documentpass1,967ms221found
Return valid JSON that matches a schemapass2,388ms398population=9776000
Answer in exactly seven wordspass2,789ms5557 words
Hold three format rules at oncepass4,199ms9313×4, capitalised

Findings

It passed everything we could ask

All 5 tasks completed. What separates it from the others here is cost, not capability.

It reports thinking tokens

2,260 across 5 tasks. Those are billed, and they land before the first output token — which is why wall time and output length disagree on this model.

4 tasks could not be asked

The free-tier quota was spent before we got to Write a sentence without the letter E, Return an array of exactly five items, Reverse a list exactly, Four-step arithmetic, number only. Those are left out of the pass rate rather than counted against the model.

How this was measured

Every task was sent at temperature 0, one request at a time, and judged by code — the checkers live in bench/tasks.mjs. Failed responses are stored verbatim so the verdict can be re-read. See all models and tasks.