DeepSeek-V3 vs Kimi K2 (Non-Reasoning)
Wins 1 of 6 areas
Coding
Wins 4 of 6 areas
Agents · Reasoning · Facts · Long documents
Kimi K2 (Non-Reasoning) is the stronger all-rounder.DeepSeek-V3 is cheaper and better at coding.
Scores updated · 19 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.
- ReasoningHard problems that need careful thinking03Kimi K2 (Non-Reasoning)3 of 4 tests · 1 tie
- FactsGetting facts right instead of making them up12Kimi K2 (Non-Reasoning)2 of 3 tests
- AgentsCarrying out multi-step tasks on its own01Kimi K2 (Non-Reasoning)1 of 1 test
- Long documentsFinding answers in very long texts01Kimi K2 (Non-Reasoning)1 of 1 test
- CodingWriting and fixing software10DeepSeek-V31 of 2 tests · 1 tie
- Following instructionsDoing exactly what it is asked00Even0 each · 1 tie
Agents, 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 60% 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 (Non-Reasoning) pulls ahead
- Reasons across sets of long documentsAA-LCR+12.3points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+11.1points ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+9.3points ahead
Where DeepSeek-V3 pulls ahead
- Code for real scientific research problemsSciCode+4.5points ahead
Every test, side by side
All 19 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3
- SciCodeDeepSeek-V3 by 4.53934.5+4.5
- Terminal-Bench Hardtie15.215.9tie
ReasoningKimi K2 (Non-Reasoning)
- GPQA DiamondKimi K2 (Non-Reasoning) by 11.165.576.6+11.1
- SimpleBenchKimi K2 (Non-Reasoning) by 7.418.926.3+7.4
- Humanity's Last ExamKimi K2 (Non-Reasoning) by 2.74.77.4+2.7
- CritPttie00tie
FactsKimi K2 (Non-Reasoning)
- Vectara HHEM hallucination ratelower is betterDeepSeek-V3 by 11.86.117.9+11.8
- AA-Omniscience · Non-hallucinationKimi K2 (Non-Reasoning) by 9.314.123.4+9.3
- AA-Omniscience · AccuracyKimi K2 (Non-Reasoning) by 1.925.427.4+1.9
Long documentsKimi K2 (Non-Reasoning)
- AA-LCRKimi K2 (Non-Reasoning) by 12.340.753+12.3
Other results7 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.
- vectara_avg_summary_lengthDeepSeek-V3 by 22.581.759.2+22.5
- τ²-Bench Telecom (AA run)Kimi K2 (Non-Reasoning) by 1447.161.1+14
- AA-OmniscienceKimi K2 (Non-Reasoning) by 12.4-40.7-28.3+12.4
- Artificial Analysis Coding IndexKimi K2 (Non-Reasoning) by 11.82334.8+11.8
- vectara_factual_consistencyDeepSeek-V3 by 11.893.982.1+11.8
- AA IntelligenceKimi K2 (Non-Reasoning) by 39.712.7+3
- vectara_answer_rateKimi K2 (Non-Reasoning) by 1.197.598.6+1.1
Questions people ask
Which is better, DeepSeek-V3 or Kimi K2 (Non-Reasoning)?
Kimi K2 (Non-Reasoning) wins four of the six areas where both have results: agents, reasoning, facts and long documents. DeepSeek-V3 wins coding, and costs 60% less. They are level on following instructions.
Which is better for coding?
DeepSeek-V3. It wins 1 of the 2 coding tests both models report; Kimi K2 (Non-Reasoning) wins none, and 1 is a tie.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Kimi K2 (Non-Reasoning) costs $0.57 and $2.30. That makes DeepSeek-V3 about 60% cheaper for the same work.
How do you compare the two?
We use the 19 benchmark tests both models have published scores on. The verdict counts the 12 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 7 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.