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

Task

Write a sentence without the letter E

A constraint the model must hold across every token it writes. The most common letter in English.

Measured
2026-09-12 UTC
Result
2 of 3 passed · 3 not asked
Temperature
0
Prompt size
112 chars
ModelResultTimeThinking tokensWhat came back
gemini-2.5-flashnot asked213msnot reported429 RESOURCE_EXHAUSTED
gemini-2.5-flash-litenot asked218msnot reported429 RESOURCE_EXHAUSTED
gemini-3.5-flashnot asked377msnot reported429 RESOURCE_EXHAUSTED
gemini-3.1-flash-litepass737msnot reported8 words, clean
gemini-3.5-flash-litefail945msnot reported1 × "e"
gemini-3-flash-previewpass13,869ms3,9306 words, clean

What happened

3 models could not be asked

gemini-2.5-flash returned 429 RESOURCE_EXHAUSTED
gemini-2.5-flash-lite returned 429 RESOURCE_EXHAUSTED
gemini-3.5-flash returned 429 RESOURCE_EXHAUSTED. That is a spent quota, not a wrong answer — it is left out of the pass rate rather than counted as a failure.

1 of 3 models failed

gemini-3.5-flash-lite — 1 × "e"

65.1× between fastest and slowest

gemini-2.5-flash finished in 213ms. gemini-3-flash-preview took 13,869ms for the same prompt.

The exact prompt

Describe the ocean in one sentence. Do not use the letter 'e' anywhere in your answer. Output only the sentence.

Pass/fail is decided by code, not by reading the answer. The check for this task is in bench/tasks.mjs, and failed responses are stored verbatim so the verdict can be re-read.