DeepSeek-V3 vs Qwen3 32B
Wins 3 of 6 areas
Coding · Long documents · Following instructions
Wins 1 of 6 areas
Reasoning
DeepSeek-V3 wins more areas, narrowly.Qwen3 32B is cheaper and better at reasoning.
Scores updated · 29 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 software30DeepSeek-V33 of 3 tests
- Long documentsFinding answers in very long texts10DeepSeek-V31 of 1 test
- Following instructionsDoing exactly what it is asked10DeepSeek-V31 of 1 test
- ReasoningHard problems that need careful thinking02Qwen3 32B2 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own00Even0 each · 3 ties
- FactsGetting facts right instead of making them up11Even1 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.
Qwen3 32B costs 30% less for the same work.
The biggest differences
The three tests each model wins by the widest margin. Scores are out of 100.
Where DeepSeek-V3 pulls ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+12.2points ahead
- Command-line tasks in a real terminalTerminal-Bench 2.1+11.7points ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+8.1points ahead
Where Qwen3 32B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+3.8points ahead
- Very hard expert questions across many subjectsHumanity's Last Exam+2.7points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+1.3points ahead
Every test, side by side
All 29 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingDeepSeek-V3
- Terminal-Bench HardDeepSeek-V3 by 12.215.23+12.2
- Terminal-Bench 2.1DeepSeek-V3 by 11.716.95.2+11.7
- SciCodeDeepSeek-V3 by 33936+3
AgentsEven
- τ-Bench V3 · Bankingtie4.75.4tie
- GDPValtie00tie
- Terminal-Bench 4.0tie00tie
ReasoningQwen3 32B
- Humanity's Last ExamQwen3 32B by 2.74.77.4+2.7
- GPQA DiamondQwen3 32B by 1.365.566.8+1.3
- CritPttie00.3tie
FactsEven
- AA-Omniscience · AccuracyDeepSeek-V3 by 8.125.417.4+8.1
- AA-Omniscience · Non-hallucinationQwen3 32B by 3.814.117.9+3.8
- Vectara HHEM hallucination ratelower is bettertie6.15.9tie
Following instructionsDeepSeek-V3
- IFBenchDeepSeek-V3 by 4.74136.3+4.7
Other results15 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.
- AIME 2024Qwen3 32B by 42.239.281.4+42.2
- AIME 2025Qwen3 32B by 21.651.372.9+21.6
- τ²-Bench Telecom (AA run)DeepSeek-V3 by 17.347.129.8+17.3
- LiveCodeBenchQwen3 32B by 16.549.265.7+16.5
- MMLU-ProXDeepSeek-V3 by 10.670.559.9+10.6
- AA-OmniscienceDeepSeek-V3 by 9.7-40.7-50.4+9.7
- Aider-PolyglotDeepSeek-V3 by 9.649.640+9.6
- Artificial Analysis Coding IndexDeepSeek-V3 by 7.72315.3+7.7
- MMLU-ProQwen3 32B by 5.975.981.8+5.9
- vectara_avg_summary_lengthQwen3 32B by 34.1 rating points81.7115.8+34.1 rating
- LiveBenchQwen3 32B by 2.572.474.9+2.5
- Arena HardQwen3 32B by 2.491.493.8+2.4
- vectara_answer_rateQwen3 32B by 2.497.599.9+2.4
- AA IntelligenceDeepSeek-V3 by 1.19.78.6+1.1
- vectara_factual_consistencytie93.994.1tie
Questions people ask
Which is better, DeepSeek-V3 or Qwen3 32B?
DeepSeek-V3 wins three of the six areas where both have results: coding, long documents and following instructions. Qwen3 32B wins reasoning, and costs 30% less. They are level on agents and facts.
Which is better for coding?
DeepSeek-V3. It wins 3 of the 3 coding tests both models report; Qwen3 32B wins none.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Qwen3 32B costs $0.16 and $0.64. That makes Qwen3 32B about 30% cheaper for the same work.
How do you compare the two?
We use the 29 benchmark tests both models have published scores on. The verdict counts the 14 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 15 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.