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

Model

gemini-2.5-flash

4 of 5 tasks passed, in 10,305 ms of wall time and 1,104 thinking tokens.

Tasks passed
4 of 5
Total time
10,305 ms
Thinking tokens
1,104
One-word probe
1,139 ms
Not asked
4 (quota)
TaskResultTime Thinking tokensWhat came back
Output the number and nothing elsepass1,315ms118clean
Return valid JSON that matches a schemapass2,107ms252population=9700000
Find one fact buried in a long documentpass2,225ms81found
Hold three format rules at oncepass2,262ms3263×4, capitalised
Answer in exactly seven wordsfail2,396ms3276 words

Findings

Where it failed

Answer in exactly seven words — 6 words

It reports thinking tokens

1,104 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.