Confirm Action

Are you sure you want to proceed?

Google · 7 models tested

Which Google model should you use?

Google's strongest model is Gemini 3.1 Pro Preview (avg percentile 84, top-3 in 3 of 22 task areas). On a budget, Gemini 3.5 Flash stays close at $10.5/1M.

The Google lineup, ranked

Model Overall Code & dataWriting & commsBusiness & strategyCreative & visual AA Intel† Top-3s Price / 1M
Gemini 3.1 Pro Preview Best overall
22 task areas
84 76 90 82 91 46 3 $14.0
Gemini 3.5 Flash Best value
22 task areas
75 78 80 65 87 50 0 $10.5
Gemini 3.1 Flash Lite Preview
4 task areas
67 61 86 25 0 $1.75
Gemini 3 Flash Preview
22 task areas
64 75 64 51 80 2 $3.5
Gemini 3.1 Flash Lite
22 task areas
58 61 70 45 64 0 $1.75
Gemini 2.5 Pro
1 task areas
37 37 26 0 $11.25
Gemini 2.5 Flash
1 task areas
10 10 0 $0.75

Area figures are mean percentile ranks; 85+ 70–84 55–69 40–54 <40 · ★ = the lineup's best in that area. †AA Intel = Artificial Analysis Intelligence Index, third-party reference only.

Best Google model, task by task

The lineup's strongest model in each task area — and where it lands against every model we test, not just Google's. #1 means it beats every rival lab too.

Where another lab clearly wins

Honesty corner: the task areas where even Google's best model ranks furthest from the top. See who actually leads there.

Frequently asked

What is the best Google model?

Gemini 3.1 Pro Preview — the highest average percentile (84) across the task areas we benchmarked, with 3 top-3 finishes.

What is the best cheap Google model?

Gemini 3.5 Flash: within 9 percentile points of Gemini 3.1 Pro Preview at $10.5 per 1M tokens.

Where does Google lead every other lab?

A Google model holds the outright #1 spot in 2 task areas: Customer Support, Frontend & Landing Pages.

Every model, every lab — full leaderboard → Best model by task →

This page is Spring Prompt, running

We just did this for every model. Do it for your prompt.

The rankings above come from running real tasks through real models and scoring every output. Spring Prompt is that same engine — pointed at your prompt, your test cases, and your definition of good.

  • Generate test cases from your prompt — no eval set required to start.
  • Compare models side by side with quality, cost and latency in one matrix.
  • Optimise the winner until the scores say it's ready to ship.
Experiment · Cold outreach email

Prompt × model results

12 test cases · 3 evals
Claude Opus
GPT-5
Gemini
v1
7.1
6.8
7.4
v2
8.3
7.9
8.0
v3
9.2
8.6
8.4
Best combo: v3 × Claude Opus
9.2 quality · $0.004/run · 1.8s