Claude Sonnet 4.6 vs Gemini 3.5 Flash
Wins 2 of 8 areas
Coding · Long documents
Wins 6 of 8 areas
Agents · Reasoning · Facts · Images and charts · Math · Following instructions
Gemini 3.5 Flash is the stronger all-rounder.Claude Sonnet 4.6 is better at long documents.
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.
- ReasoningHard problems that need careful thinking14Gemini 3.5 Flash4 of 5 tests
- Images and chartsUnderstanding pictures, charts and video03Gemini 3.5 Flash3 of 3 tests
- AgentsCarrying out multi-step tasks on its own24Gemini 3.5 Flash4 of 7 tests · 1 tie
- Following instructionsDoing exactly what it is asked02Gemini 3.5 Flash2 of 2 tests
- FactsGetting facts right instead of making them up12Gemini 3.5 Flash2 of 3 tests
- MathCompetition and research-level math01Gemini 3.5 Flash1 of 1 test
- Long documentsFinding answers in very long texts20Claude Sonnet 4.62 of 2 tests
- CodingWriting and fixing software43Claude Sonnet 4.64 of 7 tests
Math 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 3.5 Flash costs 42% 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 3.5 Flash pulls ahead
- Short factual questions, answered correctlySimpleQA Verified+33.4points ahead
- Multi-step tasks using many real software toolsMCP Atlas+22.3points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+19.7points ahead
Where Claude Sonnet 4.6 pulls ahead
- Finds one of eight look-alike replies in a long chatMRCR v2 8 needle 128k (average)+58.3points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+13.8points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+12.1points 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.
CodingClaude Sonnet 4.6
- Terminal-Bench HardClaude Sonnet 4.6 by 12.15340.9+12.1
- Terminal-Bench 2.1Gemini 3.5 Flash by 7.571.278.7+7.5
- LiveBench · Agentic CodingGemini 3.5 Flash by 6.442.649+6.4
- SciCodeGemini 3.5 Flash by 3.850.153.9+3.8
- SWE-bench ProClaude Sonnet 4.6 by 358.155.1+3
- LMArena · WebDevClaude Sonnet 4.6 by 23 rating points15221499+23 rating
- LiveBench · CodingClaude Sonnet 4.6 by 1.179.378.2+1.1
AgentsGemini 3.5 Flash
- MCP AtlasGemini 3.5 Flash by 22.361.383.6+22.3
- AA ApexAgentsGemini 3.5 Flash by 19.12847.1+19.1
- OSWorld-VerifiedGemini 3.5 Flash by 5.972.578.4+5.9
- Terminal-Bench 4.0Gemini 3.5 Flash by 3.636.6+3.6
- τ-Bench V3 · BankingClaude Sonnet 4.6 by 2.234.432.2+2.2
- GDPValClaude Sonnet 4.6 by 1.536.735.2+1.5
- AA IT-Bench SREtie39.840.3tie
ReasoningGemini 3.5 Flash
- ARC-AGI-2Gemini 3.5 Flash by 13.858.372.1+13.8
- CritPtGemini 3.5 Flash by 103.113.1+10
- Humanity's Last ExamGemini 3.5 Flash by 9.133.642.7+9.1
- GPQA DiamondGemini 3.5 Flash by 4.787.592.2+4.7
- LiveBench · ReasoningClaude Sonnet 4.6 by 2.884.882+2.8
FactsGemini 3.5 Flash
- SimpleQA VerifiedGemini 3.5 Flash by 33.432.866.2+33.4
- AA-Omniscience · Non-hallucinationClaude Sonnet 4.6 by 13.851.637.8+13.8
- AA-Omniscience · AccuracyGemini 3.5 Flash by 10.540.951.4+10.5
Images and chartsGemini 3.5 Flash
- CharXiv (RQ)Gemini 3.5 Flash by 12.671.684.2+12.6
- MMMU-ProGemini 3.5 Flash by 1173.384.3+11
- LMArena · VisionGemini 3.5 Flash by 27 rating points12831310+27 rating
Long documentsClaude Sonnet 4.6
- MRCR v2 8 needle 128k (average)Claude Sonnet 4.6 by 58.384.926.6+58.3
- AA-LCRClaude Sonnet 4.6 by 6.78073.3+6.7
Following instructionsGemini 3.5 Flash
- IFBenchGemini 3.5 Flash by 19.756.676.3+19.7
- LiveBench · Instruction FollowingGemini 3.5 Flash by 12.463.275.6+12.4
Other results22 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.
- Blueprint-Bench 2Gemini 3.5 Flash by 26.96.733.6+26.9
- τ²-Bench Telecom (AA run)Gemini 3.5 Flash by 19.675.795.3+19.6
- Terminal-Bench 2.0Gemini 3.5 Flash by 17.159.176.2+17.1
- ToolathlonGemini 3.5 Flash by 15.54156.5+15.5
- livebench_data_analysisClaude Sonnet 4.6 by 13.17864.9+13.1
- AA-OmniscienceGemini 3.5 Flash by 912.221.2+9
- livebench_languageGemini 3.5 Flash by 8.576.184.6+8.5
- Artificial Analysis Coding IndexGemini 3.5 Flash by 7.16370.1+7.1
- DeepSWE 1.1Gemini 3.5 Flash by 73037+7
- Humanity’s Last ExamGemini 3.5 Flash by 733.240.2+7
- Finance Agent v2Gemini 3.5 Flash by 6.95157.9+6.9
- ARC-AGI-1Gemini 3.5 Flash by 6.58692.5+6.5
- frontiermath_tier_4_v1Gemini 3.5 Flash by 6.38.314.6+6.3
- AA Agentic IndexClaude Sonnet 4.6 by 5.833.127.3+5.8
- Finance AgentClaude Sonnet 4.6 by 5.463.357.9+5.4
- GDPval-AA v2Claude Sonnet 4.6 by 46 rating points13951349+46 rating
- Legal Agent BenchmarkClaude Sonnet 4.6 by 4.65.40.8+4.6
- Legal Agent Benchmark (Harvey's Held-Out Set)Claude Sonnet 4.6 by 4.65.40.8+4.6
- AA IntelligenceGemini 3.5 Flash by 2.530.132.6+2.5
- GDPval-AA (Elo)Claude Sonnet 4.6 by 20 rating points16761656+20 rating
- LiveBenchtie75.575tie
- CharXiv Reasoning (With tools)tie85.384.9tie
Questions people ask
Which is better, Claude Sonnet 4.6 or Gemini 3.5 Flash?
Gemini 3.5 Flash wins six of the eight areas we test: agents, reasoning, facts, images and charts, math and following instructions. Claude Sonnet 4.6 wins coding and long documents.
Which is better for coding?
Claude Sonnet 4.6. It wins 4 of the 7 coding tests both models report; Gemini 3.5 Flash wins 3.
Which is cheaper?
Claude Sonnet 4.6 costs $3.00 per million input tokens and $15.00 per million output tokens; Gemini 3.5 Flash costs $1.50 and $9.00. That makes Gemini 3.5 Flash about 42% 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 30 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 22 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.