DeepSeek-V3 vs Qwen3.6 35B A3B
Wins 0 of 6 areas
—
Wins 5 of 6 areas
Coding · Agents · Reasoning · Long documents · Following instructions
Qwen3.6 35B A3B is the stronger all-rounder.DeepSeek-V3 is cheaper.
Scores updated · 25 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 software14Qwen3.6 35B A3B4 of 5 tests
- AgentsCarrying out multi-step tasks on its own02Qwen3.6 35B A3B2 of 3 tests · 1 tie
- ReasoningHard problems that need careful thinking02Qwen3.6 35B A3B2 of 3 tests · 1 tie
- Long documentsFinding answers in very long texts01Qwen3.6 35B A3B1 of 1 test
- Following instructionsDoing exactly what it is asked01Qwen3.6 35B A3B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
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 57% 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 Qwen3.6 35B A3B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+35.4points ahead
- Recent programming contest problemsLiveCodeBench v6+33.5points ahead
- Fixes real GitHub issues in Python projectsSWE-bench Verified+31.4points ahead
Where DeepSeek-V3 pulls ahead
- Answers hard knowledge questions correctlyAA-Omniscience · Accuracy+6.7points ahead
- Code for real scientific research problemsSciCode+2.4points ahead
Every test, side by side
All 25 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.6 35B A3B
- LiveCodeBench v6Qwen3.6 35B A3B by 33.546.980.4+33.5
- SWE-bench VerifiedQwen3.6 35B A3B by 31.44273.4+31.4
- Terminal-Bench 2.1Qwen3.6 35B A3B by 2816.944.9+28
- Terminal-Bench HardQwen3.6 35B A3B by 19.615.234.8+19.6
- SciCodeDeepSeek-V3 by 2.43936.6+2.4
AgentsQwen3.6 35B A3B
- GDPValQwen3.6 35B A3B by 19.8019.8+19.8
- τ-Bench V3 · BankingQwen3.6 35B A3B by 4.64.79.3+4.6
- Terminal-Bench 4.0tie00tie
ReasoningQwen3.6 35B A3B
- GPQA DiamondQwen3.6 35B A3B by 18.665.584.1+18.6
- Humanity's Last ExamQwen3.6 35B A3B by 17.54.722.2+17.5
- CritPttie00.3tie
FactsEven
- AA-Omniscience · Non-hallucinationQwen3.6 35B A3B by 35.414.149.5+35.4
- AA-Omniscience · AccuracyDeepSeek-V3 by 6.725.418.8+6.7
Following instructionsQwen3.6 35B A3B
- IFBenchQwen3.6 35B A3B by 23.44164.4+23.4
Other results10 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.
- HMMT 2025Qwen3.6 35B A3B by 61.627.589.1+61.6
- HMMT Feb. 2025Qwen3.6 35B A3B by 61.529.290.7+61.5
- τ²-Bench Telecom (AA run)Qwen3.6 35B A3B by 48.247.195.3+48.2
- Artificial Analysis Coding IndexQwen3.6 35B A3B by 18.92341.9+18.9
- AA-OmniscienceQwen3.6 35B A3B by 18.5-40.7-22.2+18.5
- SuperGPQAQwen3.6 35B A3B by 1153.764.7+11
- MMLU-ProQwen3.6 35B A3B by 9.375.985.2+9.3
- AA IntelligenceQwen3.6 35B A3B by 8.59.718.2+8.5
- MMLU-ReduxQwen3.6 35B A3B by 4.289.193.3+4.2
- C-EvalQwen3.6 35B A3B by 3.586.590+3.5
Questions people ask
Which is better, DeepSeek-V3 or Qwen3.6 35B A3B?
Qwen3.6 35B A3B wins five of the six areas where both have results: coding, agents, reasoning, long documents and following instructions. DeepSeek-V3 wins none, but costs 57% less. They are level on facts.
Which is better for coding?
Qwen3.6 35B A3B. It wins 4 of the 5 coding tests both models report; DeepSeek-V3 wins 1.
Which is cheaper?
DeepSeek-V3 costs $0.24 per million input tokens and $0.90 per million output tokens; Qwen3.6 35B A3B costs $0.38 and $2.25. That makes DeepSeek-V3 about 57% cheaper for the same work.
How do you compare the two?
We use the 25 benchmark tests both models have published scores on. The verdict counts the 15 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 10 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.