DeepSeek-V3 vs Kimi K2.6
Wins 0 of 6 areas
—
Wins 6 of 6 areas
Coding · Agents · Reasoning · Facts · Long documents · Following instructions
Kimi K2.6 is the stronger all-rounder.DeepSeek-V3 is cheaper.
Scores updated · 28 tests both models report · How we compare
Where each one wins
Tests won in each of the six 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.
- CodingWriting and fixing software05Kimi K2.65 of 5 tests
- ReasoningHard problems that need careful thinking03Kimi K2.63 of 3 tests
- AgentsCarrying out multi-step tasks on its own02Kimi K2.62 of 3 tests · 1 tie
- FactsGetting facts right instead of making them up12Kimi K2.62 of 3 tests
- Long documentsFinding answers in very long texts01Kimi K2.61 of 1 test
- Following instructionsDoing exactly what it is asked01Kimi K2.61 of 1 test
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-V3 costs 77% less for the same work.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Kimi K2.6 pulls ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+49points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+45.4points ahead
- Recent programming contest problemsLiveCodeBench v6+42.7points ahead
Where DeepSeek-V3 pulls ahead
No clear win on a test scored out of 100.
Every test, side by side
All 28 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingKimi K2.6
- Terminal-Bench 2.1Kimi K2.6 by 4916.965.9+49
- LiveCodeBench v6Kimi K2.6 by 42.746.989.6+42.7
- SWE-bench VerifiedKimi K2.6 by 38.24280.2+38.2
- Terminal-Bench HardKimi K2.6 by 28.715.243.9+28.7
- SciCodeKimi K2.6 by 12.53951.5+12.5
AgentsKimi K2.6
- GDPValKimi K2.6 by 27027+27
- τ-Bench V3 · BankingKimi K2.6 by 18.64.723.3+18.6
- Terminal-Bench 4.0tie00.5tie
ReasoningKimi K2.6
- Humanity's Last ExamKimi K2.6 by 32.84.737.5+32.8
- GPQA DiamondKimi K2.6 by 25.665.591.1+25.6
- CritPtKimi K2.6 by 808+8
FactsKimi K2.6
- AA-Omniscience · Non-hallucinationKimi K2.6 by 45.414.159.5+45.4
- AA-Omniscience · AccuracyKimi K2.6 by 7.125.432.6+7.1
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3 by 4.76.110.8+4.7
Other results12 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.
- τ²-Bench Telecom (AA run)Kimi K2.6 by 48.847.195.9+48.8
- AA-OmniscienceKimi K2.6 by 46-40.75.3+46
- LiveCodeBenchKimi K2.6 by 40.449.289.6+40.4
- Artificial Analysis Coding IndexKimi K2.6 by 38.82361.8+38.8
- OJBenchKimi K2.6 by 36.62460.6+36.6
- AA IntelligenceKimi K2.6 by 17.39.727+17.3
- MMLU-ProKimi K2.6 by 11.275.987.1+11.2
- Chinese SimpleQA (C-SimpleQA)Kimi K2.6 by 7.96875.9+7.9
- vectara_factual_consistencyDeepSeek-V3 by 4.793.989.2+4.7
- vectara_avg_summary_lengthKimi K2.6 by 35 rating points81.7116.7+35 rating
- vectara_answer_rateKimi K2.6 by 2.297.599.7+2.2
- LiveBenchtie72.472.2tie
Questions people ask
Which is better, DeepSeek-V3 or Kimi K2.6?
Kimi K2.6 wins all six areas where both have results: coding, agents, reasoning, facts, long documents and following instructions. DeepSeek-V3 wins none, but costs 77% less.
Which is better for coding?
Kimi K2.6. It wins 5 of the 5 coding tests both models report; DeepSeek-V3 wins none.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Kimi K2.6 costs $0.95 and $4.00. That makes DeepSeek-V3 about 77% cheaper for the same work.
How do you compare the two?
We use the 28 benchmark tests both models have published scores on. The verdict counts the 16 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 12 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.