Devstral 2 vs Gemini 3.1 Pro
Wins 0 of 6 areas
—
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Gemini 3.1 Pro is the stronger all-rounder.Devstral 2 is cheaper.
Scores updated · 22 tests both models report · How we compare
Where each one wins
Tests won in each of the six areas where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
- CodingWriting and fixing software06Gemini 3.1 Pro6 of 6 tests
- AgentsCarrying out multi-step tasks on its own03Gemini 3.1 Pro3 of 3 tests
- ReasoningHard problems that need careful thinking03Gemini 3.1 Pro3 of 3 tests
- FactsGetting facts right instead of making them up02Gemini 3.1 Pro2 of 2 tests
- Long documentsFinding answers in very long texts01Gemini 3.1 Pro1 of 1 test
- Following instructionsDoing exactly what it is asked01Gemini 3.1 Pro1 of 1 test
Long documents and following instructions rest on a single test each.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
Devstral 2 costs 83% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Gemini 3.1 Pro pulls ahead
- Reasons across sets of long documentsAA-LCR+49.7points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+43.5points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+43.4points ahead
Where Devstral 2 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 22 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemini 3.1 Pro
- Terminal-Bench 2.1Gemini 3.1 Pro by 43.530.373.8+43.5
- Terminal-Bench HardGemini 3.1 Pro by 34.918.953.8+34.9
- SWE-bench VerifiedGemini 3.1 Pro by 26.853.880.6+26.8
- SciCodeGemini 3.1 Pro by 25.932.858.7+25.9
- LMArena · WebDevGemini 3.1 Pro by 252 rating points11941446+252 rating
- SWE-bench MultilingualGemini 3.1 Pro by 15.661.376.9+15.6
AgentsGemini 3.1 Pro
- GDPValGemini 3.1 Pro by 12.32.414.7+12.3
- τ-Bench V3 · BankingGemini 3.1 Pro by 10.910.521.4+10.9
- Terminal-Bench 4.0Gemini 3.1 Pro by 404+4
ReasoningGemini 3.1 Pro
- Humanity's Last ExamGemini 3.1 Pro by 43.43.647+43.4
- GPQA DiamondGemini 3.1 Pro by 34.759.494.1+34.7
- CritPtGemini 3.1 Pro by 17.7017.7+17.7
FactsGemini 3.1 Pro
- AA-Omniscience · Non-hallucinationGemini 3.1 Pro by 34.414.749.1+34.4
- AA-Omniscience · AccuracyGemini 3.1 Pro by 34.120.854.9+34.1
Following instructionsGemini 3.1 Pro
- IFBenchGemini 3.1 Pro by 3938.177.1+39
Other results6 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- AA-OmniscienceGemini 3.1 Pro by 78.6-46.731.9+78.6
- τ²-Bench Telecom (AA run)Gemini 3.1 Pro by 70.724.995.6+70.7
- Artificial Analysis Coding IndexGemini 3.1 Pro by 37.531.368.8+37.5
- Terminal-Bench 2.0Gemini 3.1 Pro by 35.932.668.5+35.9
- AA IntelligenceGemini 3.1 Pro by 21.18.629.7+21.1
- AA Agentic IndexGemini 3.1 Pro by 5.44.910.3+5.4
Questions people ask
Which is better, Devstral 2 or Gemini 3.1 Pro?
Gemini 3.1 Pro wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. Devstral 2 wins none, but costs 83% less.
Which is better for coding?
Gemini 3.1 Pro. It wins 6 of the 6 coding tests both models report; Devstral 2 wins none.
Which is cheaper?
Devstral 2 costs $0.40 per million input tokens and $2.00 per million output tokens; Gemini 3.1 Pro costs $2.00 and $12.00. That makes Devstral 2 about 83% cheaper for the same work.
How do you compare the two?
We use the 22 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 6 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.