Claude Opus 4.6 vs Gemini 3 Flash Preview
Wins 5 of 8 areas
Coding · Agents · Reasoning · Math · Long documents
Wins 2 of 8 areas
Images and charts · Following instructions
Claude Opus 4.6 is the stronger all-rounder.Gemini 3 Flash Preview is cheaper and better at images and charts.
Scores updated · 49 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 software50Claude Opus 4.65 of 5 tests
- ReasoningHard problems that need careful thinking40Claude Opus 4.64 of 5 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own40Claude Opus 4.64 of 4 tests
- MathCompetition and research-level math30Claude Opus 4.63 of 4 tests · 1 tie
- Long documentsFinding answers in very long texts10Claude Opus 4.61 of 2 tests · 1 tie
- Images and chartsUnderstanding pictures, charts and video12Gemini 3 Flash Preview2 of 3 tests
- Following instructionsDoing exactly what it is asked01Gemini 3 Flash Preview1 of 1 test
- FactsGetting facts right instead of making them up22Even2 each
Following instructions 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 Flash Preview costs 88% 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 Opus 4.6 pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+35.2points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+30.2points ahead
- Real work tasks from 44 professionsGDPVal+20.7points ahead
Where Gemini 3 Flash Preview pulls ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+24.9points ahead
- Short factual questions, answered correctlySimpleQA Verified+19.8points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+6.4points ahead
Every test, side by side
All 49 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingClaude Opus 4.6
- LMArena · WebDevClaude Opus 4.6 by 107 rating points15461439+107 rating
- Terminal-Bench HardClaude Opus 4.6 by 7.646.238.6+7.6
- SWE-bench ProClaude Opus 4.6 by 3.853.449.6+3.8
- SWE-bench VerifiedClaude Opus 4.6 by 2.880.878+2.8
- SciCodeClaude Opus 4.6 by 1.351.950.6+1.3
AgentsClaude Opus 4.6
- GDPValClaude Opus 4.6 by 20.755.935.2+20.7
- OSWorld-VerifiedClaude Opus 4.6 by 7.672.765.1+7.6
- AA ApexAgentsClaude Opus 4.6 by 5.33327.7+5.3
- MCP AtlasClaude Opus 4.6 by 5.362.757.4+5.3
ReasoningClaude Opus 4.6
- ARC-AGI-2Claude Opus 4.6 by 35.268.833.6+35.2
- SimpleBenchClaude Opus 4.6 by 6.567.661.1+6.5
- CritPtClaude Opus 4.6 by 412.68.6+4
- Humanity's Last ExamClaude Opus 4.6 by 3.339.936.6+3.3
- GPQA Diamondtie89.689.8tie
FactsEven
- AA-Omniscience · Non-hallucinationClaude Opus 4.6 by 30.237.27+30.2
- SimpleQA VerifiedGemini 3 Flash Preview by 19.84766.8+19.8
- AA-Omniscience · AccuracyGemini 3 Flash Preview by 6.44753.4+6.4
- Vectara HHEM hallucination ratelower is betterClaude Opus 4.6 by 1.312.213.5+1.3
Images and chartsGemini 3 Flash Preview
- MMMU-ProGemini 3 Flash Preview by 4.575.479.9+4.5
- LMArena · VisionClaude Opus 4.6 by 31 rating points13161285+31 rating
- CharXiv (RQ)Gemini 3 Flash Preview by 2.977.480.3+2.9
MathClaude Opus 4.6
- FrontierMath Tiers 1-3 (v2)Claude Opus 4.6 by 14.86651.2+14.8
- FrontierMath Tier 4Claude Opus 4.6 by 9.726.817.1+9.7
- HMMT Feb. 2026Claude Opus 4.6 by 6.896.289.4+6.8
- AIME 2026tie96.796.7tie
Long documentsClaude Opus 4.6
- MRCR v2 8 needle 128k (average)Claude Opus 4.6 by 16.88467.2+16.8
- AA-LCRtie7878tie
Following instructionsGemini 3 Flash Preview
- IFBenchGemini 3 Flash Preview by 24.953.178+24.9
Other results21 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.
- Vending-Bench 2Claude Opus 4.6 by 438259 rating points801759363500+438259 rating
- GDPval-AA (Elo)Claude Opus 4.6 by 415 rating points16191204+415 rating
- frontiermath_tier_4_v1Claude Opus 4.6 by 18.722.94.2+18.7
- AA Agentic IndexClaude Opus 4.6 by 17.967.649.7+17.9
- Terminal-Bench 2.0Claude Opus 4.6 by 17.865.447.6+17.8
- matharena_visual_math_overallGemini 3 Flash Preview by 13.572.385.8+13.5
- τ²-Bench Telecom (AA run)Claude Opus 4.6 by 11.792.180.4+11.7
- ScreenSpot-Pro (No tools)Gemini 3 Flash Preview by 11.457.769.1+11.4
- ARC-AGI-1Claude Opus 4.6 by 8.39384.7+8.3
- ToolathlonClaude Opus 4.6 by 7.456.849.4+7.4
- AA IntelligenceClaude Opus 4.6 by 5.631.926.3+5.6
- Artificial Analysis Coding IndexClaude Opus 4.6 by 5.548.142.6+5.5
- vectara_avg_summary_lengthClaude Opus 4.6 by 47.4 rating points137.690.2+47.4 rating
- Legal Agent BenchmarkClaude Opus 4.6 by 4.24.20+4.2
- LiveBenchClaude Opus 4.6 by 3.976.372.4+3.9
- AA-OmniscienceClaude Opus 4.6 by 3.613.710.1+3.6
- HMMT Nov. 2025Claude Opus 4.6 by 396.393.3+3
- vectara_factual_consistencyClaude Opus 4.6 by 1.387.886.5+1.3
- MMMLUtie91.191.8tie
- AIME 2025tie99.899.7tie
- vectara_answer_ratetie99.899.8tie
Questions people ask
Which is better, Claude Opus 4.6 or Gemini 3 Flash Preview?
Claude Opus 4.6 wins five of the eight areas we test: coding, agents, reasoning, math and long documents. Gemini 3 Flash Preview wins images and charts and following instructions, and costs 88% less. They are level on facts.
Which is better for coding?
Claude Opus 4.6. It wins 5 of the 5 coding tests both models report; Gemini 3 Flash Preview wins none.
Which is cheaper?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; Gemini 3 Flash Preview costs $0.50 and $3.00. That makes Gemini 3 Flash Preview about 88% cheaper for the same work.
How do you compare the two?
We use the 49 benchmark tests both models have published scores on. The verdict counts the 28 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 21 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.