Claude Sonnet 4 vs Gemini 2.5 Pro
Wins 4 of 7 areas
Coding · Agents · Long documents · Following instructions
Wins 3 of 7 areas
Reasoning · Facts · Images and charts
Claude Sonnet 4 wins more areas, narrowly.Gemini 2.5 Pro is cheaper and better at reasoning.
Scores updated · 60 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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.
- AgentsCarrying out multi-step tasks on its own30Claude Sonnet 43 of 3 tests
- CodingWriting and fixing software31Claude Sonnet 43 of 4 tests
- Following instructionsDoing exactly what it is asked20Claude Sonnet 42 of 2 tests
- Long documentsFinding answers in very long texts10Claude Sonnet 41 of 1 test
- ReasoningHard problems that need careful thinking14Gemini 2.5 Pro4 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03Gemini 2.5 Pro3 of 3 tests
- FactsGetting facts right instead of making them up12Gemini 2.5 Pro2 of 3 tests
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 2.5 Pro costs 38% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Claude Sonnet 4 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+61.8points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+9.5points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+7.8points ahead
Where Gemini 2.5 Pro pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+16.4points ahead
- Harder college exam questions with imagesMMMU-Pro+13.1points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+7.3points ahead
Every test, side by side
All 60 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Sonnet 4
- SWE-bench VerifiedClaude Sonnet 4 by 9.572.763.2+9.5
- Terminal-Bench 2.1Claude Sonnet 4 by 7.836.328.5+7.8
- SciCodeGemini 2.5 Pro by 6.34046.3+6.3
- Terminal-Bench HardClaude Sonnet 4 by 4.631.126.5+4.6
AgentsClaude Sonnet 4
- GDPValClaude Sonnet 4 by 9.79.70+9.7
- τ-Bench V3 · BankingClaude Sonnet 4 by 716.79.7+7
- BrowseCompClaude Sonnet 4 by 2.312.29.9+2.3
ReasoningGemini 2.5 Pro
- Humanity's Last ExamGemini 2.5 Pro by 7.310.718+7.3
- SimpleBenchGemini 2.5 Pro by 6.145.551.6+6.1
- GPQA DiamondGemini 2.5 Pro by 5.977.783.6+5.9
- CritPtGemini 2.5 Pro by 2.30.32.6+2.3
- ARC-AGI-2Claude Sonnet 4 by 15.94.9+1
FactsGemini 2.5 Pro
- AA-Omniscience · Non-hallucinationClaude Sonnet 4 by 61.870.99.1+61.8
- AA-Omniscience · AccuracyGemini 2.5 Pro by 16.422.739+16.4
- Vectara HHEM hallucination ratelower is betterGemini 2.5 Pro by 3.310.37+3.3
Images and chartsGemini 2.5 Pro
- MMMU-ProGemini 2.5 Pro by 13.161.874.9+13.1
- LMArena · VisionGemini 2.5 Pro by 72 rating points11911263+72 rating
- MMMUGemini 2.5 Pro by 5.274.479.6+5.2
Following instructionsClaude Sonnet 4
- IFBenchClaude Sonnet 4 by 654.748.7+6
- Multi-ChallengeClaude Sonnet 4 by 3.557.153.6+3.5
Other results39 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.
- HMMT 2025Gemini 2.5 Pro by 49.815.965.7+49.8
- AIME 2024Gemini 2.5 Pro by 48.643.492+48.6
- SimpleQAGemini 2.5 Pro by 34.915.950.8+34.9
- HarmBenchClaude Sonnet 4 by 32.69865.4+32.6
- livecodebench_hardGemini 2.5 Pro by 26.832.659.4+26.8
- ZebraLogicGemini 2.5 Pro by 17.973.791.6+17.9
- AA-OmniscienceClaude Sonnet 4 by 16.50.2-16.4+16.5
- AIR-Bench 2024Claude Sonnet 4 by 14.788.373.6+14.7
- livecodebench_mediumGemini 2.5 Pro by 14.67690.6+14.6
- AA Agentic IndexClaude Sonnet 4 by 14.117.63.5+14.1
- TAU-bench (retail)Claude Sonnet 4 by 13.580.567+13.5
- AIME 2025Gemini 2.5 Pro by 12.570.583+12.5
- τ²-Bench Telecom (AA run)Claude Sonnet 4 by 10.564.654.1+10.5
- Terminal-BenchClaude Sonnet 4 by 10.235.525.3+10.2
- TAU-bench (airline)Claude Sonnet 4 by 106050+10
- Artificial Analysis Coding IndexGemini 2.5 Pro by 9.137.646.7+9.1
- τ²-BenchGemini 2.5 Pro by 8.845.254+8.8
- xbench-DeepSearchClaude Sonnet 4 by 8.664.656+8.6
- GAIA (text only)Claude Sonnet 4 by 8.168.360.2+8.1
- HLE (with tools)Gemini 2.5 Pro by 8.120.328.4+8.1
- Aider-PolyglotGemini 2.5 Pro by 5.870.776.5+5.8
- LiveCodeBenchGemini 2.5 Pro by 5.768.574.2+5.7
- MATH-500 (EM)Gemini 2.5 Pro by 4.89498.8+4.8
- frontiermath_tier_4_v1Gemini 2.5 Pro by 4.204.2+4.2
- vectara_avg_summary_lengthClaude Sonnet 4 by 39.4 rating points145.8106.4+39.4 rating
- AA IntelligenceClaude Sonnet 4 by 3.918.915+3.9
- vectara_factual_consistencyGemini 2.5 Pro by 3.389.793+3.3
- BrowseComp-ZHGemini 2.5 Pro by 3.129.132.2+3.1
- ARC-AGI-1Claude Sonnet 4 by 34037+3
- simple_safety_testsClaude Sonnet 4 by 310097+3
- MMLU-ProGemini 2.5 Pro by 2.383.786+2.3
- theagentcompanyGemini 2.5 Pro by 2.33739.3+2.3
- XSTestGemini 2.5 Pro by 2.196.698.7+2.1
- FinSearchComp-globaltie4242.6tie
- vectara_answer_ratetie98.699.1tie
- ArtifactsBenchtie57.357.7tie
- anthropic_red_teamtie99.299.5tie
- bbqtie96.696.4tie
- livecodebench_easytie98.898.8tie
Questions people ask
Which is better, Claude Sonnet 4 or Gemini 2.5 Pro?
Claude Sonnet 4 wins four of the seven areas where both have results: coding, agents, long documents and following instructions. Gemini 2.5 Pro wins reasoning, facts and images and charts, and costs 38% less.
Which is better for coding?
Claude Sonnet 4. It wins 3 of the 4 coding tests both models report; Gemini 2.5 Pro wins 1.
Which is cheaper?
Claude Sonnet 4 costs $3.00 per million input tokens and $15.00 per million output tokens; Gemini 2.5 Pro costs $1.25 and $10.00. That makes Gemini 2.5 Pro about 38% cheaper for the same work.
How do you compare the two?
We use the 60 benchmark tests both models have published scores on. The verdict counts the 21 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 39 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.