# Gemini 3.1 Pro on the Cotera agent benchmark

> Pro-tier Gemini got the answer every time. CX cost $1.48 on a benchmark Mistral nailed for $0.066.

Source: https://cotera.co/benchmarks/models/gemini-3.1-pro-preview

---

Google's Gemini 3.1 Pro Preview — the larger, reasoning-heavy sibling of Flash. Marketed as a frontier reasoning model; priced like one.

**Score:** 5/5 benchmarks passed · **Cost across the matrix:** $2.73

## Strengths
- 5/5 pass with no schema drift, no JSON-envelope leakage, no "### Output:" prefacing.
- Most disciplined Marketing run from the Gemini family: 5 tool calls, all five tools with mention counts.
- Crunchbase was $0.086 — the second-cheapest Sales pass in the matrix.

## Watch-outs
- CX cost $1.48 — more than every other 5/5 model and 22x what Mistral spent on the same task. Pro Gemini's thinking budget went somewhere; the answer didn't need it.
- Marketing was $0.666 — solidly more expensive than GPT-5.5 ($0.435) for the same correct output. You're paying for reasoning the rubric didn't ask for.
- Total spend was $2.73 — second-most-expensive 5/5 model in the matrix, behind only Sonnet and Opus. Reach for Pro when you've measured that the median Flash answer isn't good enough.