Gemini 3 Flash Preview vs Kimi K2.5
Wins 7 of 8 areas
Coding · Agents · Reasoning · Facts · Images and charts · Math · Following instructions
Wins 0 of 8 areas
—
Gemini 3 Flash Preview is the stronger all-rounder.
Scores updated · 46 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 thinking40Gemini 3 Flash Preview4 of 4 tests
- CodingWriting and fixing software41Gemini 3 Flash Preview4 of 6 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own41Gemini 3 Flash Preview4 of 5 tests
- Images and chartsUnderstanding pictures, charts and video30Gemini 3 Flash Preview3 of 3 tests
- FactsGetting facts right instead of making them up21Gemini 3 Flash Preview2 of 4 tests · 1 tie
- MathCompetition and research-level math10Gemini 3 Flash Preview1 of 2 tests · 1 tie
- Following instructionsDoing exactly what it is asked10Gemini 3 Flash Preview1 of 1 test
- Long documentsFinding answers in very long texts00Even0 each · 1 tie
Long documents and following instructions rest on a single test each.
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 3% 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 Flash Preview pulls ahead
- Short factual questions, answered correctlySimpleQA Verified+32.5points ahead
- Abstract visual puzzles that people can solveARC-AGI-2+21.8points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+18.2points ahead
Where Kimi K2.5 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+27.3points ahead
- Multi-step tasks using many real software toolsMCP Atlas+6.4points ahead
- Long, multi-file coding tasks in real codebasesSWE-bench Pro+1.1points ahead
Every test, side by side
All 46 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingGemini 3 Flash Preview
- Terminal-Bench 2.1Gemini 3 Flash Preview by 12.35845.7+12.3
- Terminal-Bench HardGemini 3 Flash Preview by 3.838.634.8+3.8
- SciCodeGemini 3 Flash Preview by 1.650.649+1.6
- SWE-bench VerifiedGemini 3 Flash Preview by 1.27876.8+1.2
- SWE-bench ProKimi K2.5 by 1.149.650.7+1.1
- LMArena · WebDevtie14391436tie
AgentsGemini 3 Flash Preview
- GDPValGemini 3 Flash Preview by 1835.217.2+18
- AA ApexAgentsGemini 3 Flash Preview by 16.227.711.5+16.2
- τ-Bench V3 · BankingGemini 3 Flash Preview by 6.620.814.2+6.6
- MCP AtlasKimi K2.5 by 6.457.463.8+6.4
- OSWorld-VerifiedGemini 3 Flash Preview by 1.865.163.3+1.8
ReasoningGemini 3 Flash Preview
- ARC-AGI-2Gemini 3 Flash Preview by 21.833.611.8+21.8
- Humanity's Last ExamGemini 3 Flash Preview by 5.936.630.7+5.9
- CritPtGemini 3 Flash Preview by 5.58.63.1+5.5
- GPQA DiamondGemini 3 Flash Preview by 1.989.887.9+1.9
FactsGemini 3 Flash Preview
- SimpleQA VerifiedGemini 3 Flash Preview by 32.566.834.3+32.5
- AA-Omniscience · Non-hallucinationKimi K2.5 by 27.3734.3+27.3
- AA-Omniscience · AccuracyGemini 3 Flash Preview by 18.253.435.2+18.2
- Vectara HHEM hallucination ratelower is bettertie13.514.2tie
Images and chartsGemini 3 Flash Preview
- MMMU-ProGemini 3 Flash Preview by 4.579.975.4+4.5
- CharXiv (RQ)Gemini 3 Flash Preview by 2.880.377.5+2.8
- LMArena · VisionGemini 3 Flash Preview by 16 rating points12851269+16 rating
MathGemini 3 Flash Preview
- HMMT Feb. 2026Gemini 3 Flash Preview by 2.389.487.1+2.3
- AIME 2026tie96.795.8tie
Following instructionsGemini 3 Flash Preview
- IFBenchGemini 3 Flash Preview by 7.87870.2+7.8
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.
- Vending-Bench 2Gemini 3 Flash Preview by 362302 rating points3635001198+362302 rating
- AA Agentic IndexGemini 3 Flash Preview by 2849.721.7+28
- ToolathlonGemini 3 Flash Preview by 21.649.427.8+21.6
- ARC-AGI-1Gemini 3 Flash Preview by 19.484.765.3+19.4
- AA-OmniscienceGemini 3 Flash Preview by 17.410.1-7.3+17.4
- τ²-Bench Telecom (AA run)Kimi K2.5 by 15.580.495.9+15.5
- vectara_answer_rateGemini 3 Flash Preview by 7.699.892.2+7.6
- matharena_visual_math_overallGemini 3 Flash Preview by 5.285.880.6+5.2
- swe_bench_bashGemini 3 Flash Preview by 575.870.8+5
- Artificial Analysis Coding IndexKimi K2.5 by 4.242.646.8+4.2
- HMMT Nov. 2025Gemini 3 Flash Preview by 4.193.389.2+4.1
- AIME 2025Gemini 3 Flash Preview by 3.699.796.1+3.6
- LiveBenchGemini 3 Flash Preview by 3.372.469.1+3.3
- Terminal-Bench 2.0Kimi K2.5 by 3.247.650.8+3.2
- AA IntelligenceGemini 3 Flash Preview by 2.826.323.5+2.8
- vectara_avg_summary_lengthKimi K2.5 by 21.8 rating points90.2112+21.8 rating
- HMMT Feb. 2025Gemini 3 Flash Preview by 2.197.595.4+2.1
- vectara_factual_consistencytie86.585.8tie
- VideoMMMUtie86.986.6tie
- frontiermath_tier_4_v1tie4.24.2tie
Questions people ask
Which is better, Gemini 3 Flash Preview or Kimi K2.5?
Gemini 3 Flash Preview wins seven of the eight areas we test: coding, agents, reasoning, facts, images and charts, math and following instructions. Kimi K2.5 wins none. They are level on long documents.
Which is better for coding?
Gemini 3 Flash Preview. It wins 4 of the 6 coding tests both models report; Kimi K2.5 wins 1, and 1 is a tie.
Which is cheaper?
Gemini 3 Flash Preview costs $0.50 per million input tokens and $3.00 per million output tokens; Kimi K2.5 costs $0.60 and $3.00. That makes Gemini 3 Flash Preview about 3% cheaper for the same work.
How do you compare the two?
We use the 46 benchmark tests both models have published scores on. The verdict counts the 26 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.