Racing Gemma 4 31B (Google) against a typical person on draft a phd research proposal: 8 hr by hand versus 32 s for the model, 914× faster, 75% accurate on this task's rubric. ✗ Too small for this one. Try GLM-5.2. It used 951 tokens for a cost of $0.00: see how that stacks up against every other model →

Gemma 4 31B vs. a person: "Draft a PhD research proposal"

Choose a task and model. Compare its recorded or estimated speed, accuracy, and cost with a typical human time.

Get your real number

Type for 15 seconds. We'll use your real speed instead of the generic average, for every writing task on this visit — nothing you type is sent anywhere, it stays in this browser.
Taskscroll or tap · ← →
Choose the Model
AnthropicOpenAIMetaDeepSeekGoogleZ-AINVIDIAMiniMaxKrutrimSarvam AI
—
Simple? — 0/25 raced
Keep going See why →

Want the head-to-head?

Run the same task on Claude and GPT side by side.

Compare AI Models →

"You" is a typical time for a person doing the task by hand, unless you've taken the 15-second typing test above, in which case it's your own. "AI" is a recorded run for the selected model, shown to scale. Stage times, accuracy, and token cost are placeholder profiles until runs are recorded and scored. How we measure → · FAQ → · Model comparison → · Open dataset →