DeepSeek-V4.1-Flash vs Kimi K3
Wins 2 of 8 areas
Agents · Math
Wins 6 of 8 areas
Coding · Reasoning · Facts · Images and charts · Long documents · Following instructions
Kimi K3 is the stronger all-rounder.DeepSeek-V4.1-Flash is cheaper and better at agents.
Scores updated · 42 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 up03Kimi K33 of 3 tests
- ReasoningHard problems that need careful thinking24Kimi K34 of 6 tests
- Images and chartsUnderstanding pictures, charts and video02Kimi K32 of 2 tests
- CodingWriting and fixing software23Kimi K33 of 5 tests
- Long documentsFinding answers in very long texts01Kimi K31 of 1 test
- Following instructionsDoing exactly what it is asked01Kimi K31 of 1 test
- AgentsCarrying out multi-step tasks on its own20DeepSeek-V4.1-Flash2 of 3 tests · 1 tie
- MathCompetition and research-level math21DeepSeek-V4.1-Flash2 of 3 tests
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.
DeepSeek-V4.1-Flash costs 92% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where Kimi K3 pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+43.3points ahead
- Reasoning about charts from research papersCharXiv (RQ)+22.3points ahead
- Research-level physics problemsCritPt+9.1points ahead
Where DeepSeek-V4.1-Flash pulls ahead
- Abstract visual puzzles that people can solveARC-AGI-2+17.7points ahead
- Fixes real GitHub issues, working as an agentLiveBench · Agentic Coding+15.1points ahead
- Complex command-line tasks across many fieldsTerminal-Bench 4.0+14.2points ahead
Every test, side by side
All 42 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K3
- LiveBench · Agentic CodingDeepSeek-V4.1-Flash by 15.177.362.2+15.1
- SciCodeKimi K3 by 7.651.959.5+7.6
- Terminal-Bench 2.1DeepSeek-V4.1-Flash by 5.690.685+5.6
- LMArena · WebDevKimi K3 by 37 rating points16191656+37 rating
- LiveBench · CodingKimi K3 by 1.58081.5+1.5
AgentsDeepSeek-V4.1-Flash
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 14.226.812.6+14.2
- GDPValDeepSeek-V4.1-Flash by 3.35551.7+3.3
- AA IT-Bench SREtie46.947.7tie
ReasoningKimi K3
- ARC-AGI-2DeepSeek-V4.1-Flash by 17.778.160.4+17.7
- CritPtKimi K3 by 9.114.323.4+9.1
- Humanity's Last ExamKimi K3 by 7.739.246.9+7.7
- SimpleBenchDeepSeek-V4.1-Flash by 666.760.7+6
- LiveBench · ReasoningKimi K3 by 486.790.7+4
- GPQA DiamondKimi K3 by 2.690.993.5+2.6
FactsKimi K3
- AA-Omniscience · Non-hallucinationKimi K3 by 43.33.546.8+43.3
- SimpleQA VerifiedKimi K3 by 7.742.950.6+7.7
- AA-Omniscience · AccuracyKimi K3 by 1.246.447.6+1.2
Images and chartsKimi K3
- CharXiv (RQ)Kimi K3 by 22.36991.3+22.3
- MMMU-ProKimi K3 by 3.57780.5+3.5
MathDeepSeek-V4.1-Flash
- LiveBench · MathematicsDeepSeek-V4.1-Flash by 8.993.384.4+8.9
- FrontierMath Tier 4DeepSeek-V4.1-Flash by 7.346.339+7.3
- FrontierMath Tiers 1-3 (v2)Kimi K3 by 171.272.2+1
Following instructionsKimi K3
- LiveBench · Instruction FollowingKimi K3 by 1.47071.4+1.4
Other results18 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.
- Program BenchKimi K3 by 57.520.377.8+57.5
- AA-OmniscienceKimi K3 by 25-5.319.7+25
- Terminal-Bench 4.0DeepSeek-V4.1-Flash by 18.631.212.6+18.6
- Terminal-bench 3.0DeepSeek-V4.1-Flash by 12.63017.4+12.6
- CyberGymDeepSeek-V4.1-Flash by 8.188.180+8.1
- ZeroBenchDeepSeek-V4.1-Flash by 84941+8
- SWE Refactor BenchKimi K3 by 7.711.819.5+7.7
- PostTrain Bench v1.1Kimi K3 by 6.130.536.6+6.1
- NL2RepoDeepSeek-V4.1-Flash by 66458+6
- DeepSWE 1.1DeepSeek-V4.1-Flash by 5.274.269+5.2
- livebench_languageKimi K3 by 4.381.285.5+4.3
- AA IntelligenceKimi K3 by 4.139.543.6+4.1
- HLE (with tools)DeepSeek-V4.1-Flash by 4.163.959.8+4.1
- OSWorld 2.0Kimi K3 by 454.358.3+4
- BabyVisionDeepSeek-V4.1-Flash by 3.989.685.7+3.9
- Agents' Last ExamDeepSeek-V4.1-Flash by 3.531.828.3+3.5
- ARC-AGI-1Kimi K3 by 193.594.5+1
- livebench_data_analysistie79.378.7tie
Questions people ask
Which is better, DeepSeek-V4.1-Flash or Kimi K3?
Kimi K3 wins six of the eight areas we test: coding, reasoning, facts, images and charts, long documents and following instructions. DeepSeek-V4.1-Flash wins agents and math, and costs 92% less.
Which is better for coding?
Kimi K3. It wins 3 of the 5 coding tests both models report; DeepSeek-V4.1-Flash wins 2.
Which is cheaper?
DeepSeek-V4.1-Flash costs $0.30 per million input tokens and $1.20 per million output tokens; Kimi K3 costs $3.00 and $15.00. That makes DeepSeek-V4.1-Flash about 92% cheaper for the same work.
How do you compare the two?
We use the 42 benchmark tests both models have published scores on. The verdict counts the 24 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 18 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.