Gemini 3.1 Pro vs GPT-5.4
Wins 4 of 8 areas
Reasoning · Facts · Images and charts · Following instructions
Wins 3 of 8 areas
Coding · Agents · Math
Gemini 3.1 Pro wins more areas, narrowly.GPT-5.4 is better at math.
Scores updated · 92 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.
- FactsGetting facts right instead of making them up31Gemini 3.1 Pro3 of 4 tests
- Following instructionsDoing exactly what it is asked20Gemini 3.1 Pro2 of 2 tests
- ReasoningHard problems that need careful thinking32Gemini 3.1 Pro3 of 6 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video10Gemini 3.1 Pro1 of 3 tests · 2 ties
- MathCompetition and research-level math07GPT-5.47 of 7 tests
- CodingWriting and fixing software26GPT-5.46 of 8 tests
- AgentsCarrying out multi-step tasks on its own34GPT-5.44 of 7 tests
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
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 20% 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
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+40.8points ahead
- Short factual questions, answered correctlySimpleQA Verified+28.4points ahead
- Follows detailed instructions exactlyLiveBench · Instruction Following+8.9points ahead
Where GPT-5.4 pulls ahead
- Real work tasks from 44 professionsGDPVal+22.7points ahead
- Extremely hard research-level math problemsFrontierMath Tier 4+22.2points ahead
- Proof problems from the US Math OlympiadUSAMO 2026+20.8points ahead
Every test, side by side
All 92 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGPT-5.4
- LiveBench · Agentic CodingGPT-5.4 by 9.744.153.8+9.7
- SWE-bench MultilingualGemini 3.1 Pro by 5.276.971.7+5.2
- Terminal-Bench 2.1GPT-5.4 by 4.573.878.3+4.5
- Terminal-Bench HardGPT-5.4 by 3.853.857.6+3.8
- SWE-bench ProGPT-5.4 by 3.554.257.7+3.5
- SciCodeGemini 3.1 Pro by 2.158.756.6+2.1
- LMArena · WebDevGPT-5.4 by 18 rating points14471465+18 rating
- LiveBench · CodingGPT-5.4 by 176.577.5+1
AgentsGPT-5.4
- GDPValGPT-5.4 by 22.714.737.4+22.7
- τ-Bench V3 · BankingGPT-5.4 by 18.221.439.6+18.2
- AA IT-Bench SREGPT-5.4 by 4.230.334.5+4.2
- BrowseCompGemini 3.1 Pro by 3.285.982.7+3.2
- MCP AtlasGemini 3.1 Pro by 269.267.2+2
- AA ApexAgentsGPT-5.4 by 1.33233.3+1.3
- OSWorld-VerifiedGemini 3.1 Pro by 1.276.275+1.2
ReasoningGemini 3.1 Pro
- CritPtGPT-5.4 by 5.717.723.4+5.7
- LiveBench · ReasoningGPT-5.4 by 4.18488.1+4.1
- ARC-AGI-2Gemini 3.1 Pro by 3.877.173.3+3.8
- Humanity's Last ExamGemini 3.1 Pro by 3.34743.7+3.3
- GPQA DiamondGemini 3.1 Pro by 2.194.192+2.1
- ARC-AGI-3tie0.40.2tie
FactsGemini 3.1 Pro
- AA-Omniscience · Non-hallucinationGemini 3.1 Pro by 40.849.18.3+40.8
- SimpleQA VerifiedGemini 3.1 Pro by 28.473.545.1+28.4
- AA-Omniscience · AccuracyGemini 3.1 Pro by 454.950.9+4
- Vectara HHEM hallucination ratelower is betterGPT-5.4 by 3.410.47+3.4
Images and chartsGemini 3.1 Pro
- MMMU-ProGemini 3.1 Pro by 482.478.4+4
- LMArena · Visiontie12961302tie
- CharXiv (RQ)tie83.382.8tie
MathGPT-5.4
- FrontierMath Tier 4GPT-5.4 by 22.226.849+22.2
- USAMO 2026GPT-5.4 by 20.874.495.2+20.8
- FrontierMath Tiers 1-3 (v2)GPT-5.4 by 18.959.678.6+18.9
- HMMT Feb. 2026GPT-5.4 by 10.487.397.7+10.4
- IMOAnswerBenchGPT-5.4 by 10.48191.4+10.4
- LiveBench · MathematicsGPT-5.4 by 3.29194.2+3.2
- AIME 2026GPT-5.4 by 198.299.2+1
Following instructionsGemini 3.1 Pro
- LiveBench · Instruction FollowingGemini 3.1 Pro by 8.979.170.2+8.9
- IFBenchGemini 3.1 Pro by 3.277.173.9+3.2
Other results54 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.
- DeepSWE 1.1GPT-5.4 by 401252+40
- GDPval-AA (Elo)GPT-5.4 by 357 rating points13171674+357 rating
- AA Agentic IndexGPT-5.4 by 33.910.344.2+33.9
- CyberGymGPT-5.4 by 27.538.866.3+27.5
- AA-OmniscienceGemini 3.1 Pro by 26.131.95.8+26.1
- HiL-Bench (Tools-allowed)Gemini 3.1 Pro by 25.635.39.7+25.6
- OfficeQA ProGemini 3.1 Pro by 21.472.551.1+21.4
- baby_vision_with_pythonGPT-5.4 by 11.968.380.2+11.9
- Codeforces (Rating)GPT-5.4 by 116 rating points30523168+116 rating
- AgentWorldBench - MCPGPT-5.4 by 1159.170.1+11
- Apex Shortlist (Pass@1)Gemini 3.1 Pro by 1189.178.1+11
- frontiermath_tier_4_v1GPT-5.4 by 10.416.727.1+10.4
- AA IntelligenceGPT-5.4 by 9.329.739+9.3
- Chinese SimpleQA (C-SimpleQA)Gemini 3.1 Pro by 9.185.976.8+9.1
- τ²-Bench Telecom (AA run)Gemini 3.1 Pro by 8.595.687.1+8.5
- NL2RepoGPT-5.4 by 7.933.441.3+7.9
- AgentWorldBench - SWEGPT-5.4 by 7.259.166.3+7.2
- AgentWorldBench - SearchGPT-5.4 by 7.130.237.3+7.1
- Apex (Pass@1)Gemini 3.1 Pro by 6.860.954.1+6.8
- MCPMarkGPT-5.4 by 6.655.962.5+6.6
- Terminal-Bench 2.0GPT-5.4 by 6.668.575.1+6.6
- Tool-DecathlonGPT-5.4 by 5.848.854.6+5.8
- ToolathlonGPT-5.4 by 5.848.854.6+5.8
- τ³-BenchGPT-5.4 by 5.867.172.9+5.8
- BabyVisionGemini 3.1 Pro by 4.754.449.7+4.7
- ARC-AGI-1Gemini 3.1 Pro by 4.39893.7+4.3
- AgentWorldBench - OverallGPT-5.4 by 3.754.658.3+3.7
- Finance AgentGemini 3.1 Pro by 3.759.756+3.7
- DeepSearchQA (accuracy)GPT-5.4 by 3.560.263.7+3.5
- MMLU-ProGemini 3.1 Pro by 3.59187.5+3.5
- vectara_factual_consistencyGPT-5.4 by 3.489.693+3.4
- DeepSearchQA (F1)Gemini 3.1 Pro by 3.381.978.6+3.3
- BrowseComp (Agent Swarm)Gemini 3.1 Pro by 3.285.982.7+3.2
- browsecomp_with_context_managerGemini 3.1 Pro by 3.285.982.7+3.2
- mmmu_pro_with_pythonGemini 3.1 Pro by 3.285.382.1+3.2
- matharena_visual_math_overallGPT-5.4 by 3.189.492.5+3.1
- livebench_languageGemini 3.1 Pro by 2.885.482.6+2.8
- vectara_avg_summary_lengthGemini 3.1 Pro by 26 rating points107.781.7+26 rating
- Artificial Analysis Coding IndexGPT-5.4 by 2.368.871.1+2.3
- MATH-VisionGPT-5.4 by 2.289.892+2.2
- MCPAtlas Public (Pass@1)Gemini 3.1 Pro by 269.267.2+2
- AgentWorldBench - OSGPT-5.4 by 1.766.968.6+1.7
- V* (w/ python)GPT-5.4 by 1.596.998.4+1.5
- AgentWorldBench - AndroidGemini 3.1 Pro by 1.461.460+1.4
- AgentWorldBench - TerminalGPT-5.4 by 1.252.553.7+1.2
- AgentWorldBench - WebGemini 3.1 Pro by 152.851.8+1
- HMMT Nov. 2025GPT-5.4 by 194.895.8+1
- livebench_data_analysistie78.579.3tie
- HLE (with tools)tie51.452tie
- vectara_answer_ratetie99.499.9tie
- Legal Agent Benchmarktie00.4tie
- LiveBenchtie79.980.3tie
- mathvision_with_pythontie95.796.1tie
- charxiv_rq_with_pythontie89.990tie
Questions people ask
Which is better, Gemini 3.1 Pro or GPT-5.4?
Gemini 3.1 Pro wins four of the eight areas we test: reasoning, facts, images and charts and following instructions. GPT-5.4 wins coding, agents and math. They are level on long documents.
Which is better for coding?
GPT-5.4. It wins 6 of the 8 coding tests both models report; Gemini 3.1 Pro wins 2.
Which is cheaper?
Gemini 3.1 Pro costs $2.00 per million input tokens and $12.00 per million output tokens; GPT-5.4 costs $2.50 and $15.00. That makes Gemini 3.1 Pro about 20% cheaper for the same work.
How do you compare the two?
We use the 92 benchmark tests both models have published scores on. The verdict counts the 38 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 54 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.