Claude Sonnet 4.5 vs Gemini 3.1 Pro
Wins 0 of 8 areas
—
Wins 8 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Long documents · Following instructions
Gemini 3.1 Pro is the stronger all-rounder.
Scores updated · 52 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. 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 software08Gemini 3.1 Pro8 of 8 tests
- ReasoningHard problems that need careful thinking05Gemini 3.1 Pro5 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03Gemini 3.1 Pro3 of 3 tests
- MathCompetition and research-level math03Gemini 3.1 Pro3 of 3 tests
- AgentsCarrying out multi-step tasks on its own24Gemini 3.1 Pro4 of 6 tests
- FactsGetting facts right instead of making them up13Gemini 3.1 Pro3 of 4 tests
- Following instructionsDoing exactly what it is asked02Gemini 3.1 Pro2 of 2 tests
- Long documentsFinding answers in very long texts01Gemini 3.1 Pro1 of 1 test
Long documents rests on a single test.
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.
Gemini 3.1 Pro costs 22% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Gemini 3.1 Pro pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+63.5points ahead
- Finds hard-to-locate facts by browsing the webBrowseComp+61.8points ahead
- Short factual questions, answered correctlySimpleQA Verified+42.8points ahead
Where Claude Sonnet 4.5 pulls ahead
- Real work tasks from 44 professionsGDPVal+5.8points ahead
- Customer-service tasks in a simulated bankτ-Bench V3 · Banking+3.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+1.8points ahead
Every test, side by side
All 52 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemini 3.1 Pro
- LiveCodeBench v6Gemini 3.1 Pro by 27.76491.7+27.7
- Terminal-Bench HardGemini 3.1 Pro by 18.235.653.8+18.2
- Terminal-Bench 2.1Gemini 3.1 Pro by 1855.873.8+18
- SciCodeGemini 3.1 Pro by 1345.758.7+13
- SWE-bench ProGemini 3.1 Pro by 10.643.654.2+10.6
- SWE-bench MultilingualGemini 3.1 Pro by 9.96776.9+9.9
- LMArena · WebDevGemini 3.1 Pro by 53 rating points13941447+53 rating
- SWE-bench VerifiedGemini 3.1 Pro by 3.477.280.6+3.4
AgentsGemini 3.1 Pro
- BrowseCompGemini 3.1 Pro by 61.824.185.9+61.8
- MCP AtlasGemini 3.1 Pro by 25.443.869.2+25.4
- OSWorld-VerifiedGemini 3.1 Pro by 14.861.476.2+14.8
- GDPValClaude Sonnet 4.5 by 5.820.514.7+5.8
- Terminal-Bench 4.0Gemini 3.1 Pro by 404+4
- τ-Bench V3 · BankingClaude Sonnet 4.5 by 3.124.521.4+3.1
ReasoningGemini 3.1 Pro
- ARC-AGI-2Gemini 3.1 Pro by 63.513.677.1+63.5
- Humanity's Last ExamGemini 3.1 Pro by 29.217.847+29.2
- SimpleBenchGemini 3.1 Pro by 25.354.379.6+25.3
- CritPtGemini 3.1 Pro by 16.61.117.7+16.6
- GPQA DiamondGemini 3.1 Pro by 10.783.494.1+10.7
FactsGemini 3.1 Pro
- SimpleQA VerifiedGemini 3.1 Pro by 42.830.773.5+42.8
- AA-Omniscience · AccuracyGemini 3.1 Pro by 2232.954.9+22
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.5 by 1.850.949.1+1.8
- Vectara HHEM hallucination ratelower is betterGemini 3.1 Pro by 1.61210.4+1.6
Images and chartsGemini 3.1 Pro
- CharXiv (RQ)Gemini 3.1 Pro by 16.167.283.3+16.1
- MMMU-ProGemini 3.1 Pro by 13.768.782.4+13.7
- MathVistaGemini 3.1 Pro by 10.479.890.2+10.4
MathGemini 3.1 Pro
- FrontierMath Tiers 1-3 (v2)Gemini 3.1 Pro by 35.823.959.6+35.8
- FrontierMath Tier 4Gemini 3.1 Pro by 24.42.426.8+24.4
- IMOAnswerBenchGemini 3.1 Pro by 15.165.981+15.1
Following instructionsGemini 3.1 Pro
- IFBenchGemini 3.1 Pro by 19.857.377.1+19.8
- Multi-ChallengeGemini 3.1 Pro by 16.155.371.4+16.1
Other results20 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.
- ARC-AGI-1Gemini 3.1 Pro by 34.363.798+34.3
- AA-OmniscienceGemini 3.1 Pro by 32-0.131.9+32
- LiveCodeBenchGemini 3.1 Pro by 20.77191.7+20.7
- HLE (with tools)Gemini 3.1 Pro by 19.43251.4+19.4
- Terminal-Bench 2.0Gemini 3.1 Pro by 18.55068.5+18.5
- τ²-Bench Telecom (AA run)Gemini 3.1 Pro by 17.578.195.6+17.5
- Artificial Analysis Coding IndexGemini 3.1 Pro by 16.752.168.8+16.7
- RealWorldQAGemini 3.1 Pro by 15.170.385.4+15.1
- matharena_visual_math_overallGemini 3.1 Pro by 13.675.889.4+13.6
- HMMT Nov. 2025Gemini 3.1 Pro by 13.181.794.8+13.1
- frontiermath_tier_4_v1Gemini 3.1 Pro by 12.54.216.7+12.5
- SimpleVQAGemini 3.1 Pro by 12.357.669.9+12.3
- AA IntelligenceGemini 3.1 Pro by 920.729.7+9
- ToolathlonGemini 3.1 Pro by 7.84148.8+7.8
- AA Agentic IndexClaude Sonnet 4.5 by 7.217.510.3+7.2
- vectara_answer_rateGemini 3.1 Pro by 3.895.699.4+3.8
- MMLU-ProGemini 3.1 Pro by 3.587.591+3.5
- MMMLUGemini 3.1 Pro by 3.589.192.6+3.5
- vectara_avg_summary_lengthClaude Sonnet 4.5 by 20.1 rating points127.8107.7+20.1 rating
- vectara_factual_consistencyGemini 3.1 Pro by 1.68889.6+1.6
Questions people ask
Which is better, Claude Sonnet 4.5 or Gemini 3.1 Pro?
Gemini 3.1 Pro wins all eight areas we test: coding, agents, reasoning, facts, images and charts, math, long documents and following instructions. Claude Sonnet 4.5 wins none.
Which is better for coding?
Gemini 3.1 Pro. It wins 8 of the 8 coding tests both models report; Claude Sonnet 4.5 wins none.
Which is cheaper?
Claude Sonnet 4.5 costs $3.00 per million input tokens and $15.00 per million output tokens; Gemini 3.1 Pro costs $2.00 and $12.00. That makes Gemini 3.1 Pro about 22% cheaper for the same work.
How do you compare the two?
We use the 52 benchmark tests both models have published scores on. The verdict counts the 32 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 20 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.