Qwen3.5 2B vs Qwen3.5 35B A3B
Wins 0 of 7 areas
—
Wins 6 of 7 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
Qwen3.5 35B A3B is the stronger all-rounder.
Scores updated · 36 tests both models report · How we compare
Where each one wins
Tests won in each of the seven 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.
- Images and chartsUnderstanding pictures, charts and video04Qwen3.5 35B A3B4 of 4 tests
- CodingWriting and fixing software03Qwen3.5 35B A3B3 of 3 tests
- ReasoningHard problems that need careful thinking02Qwen3.5 35B A3B2 of 3 tests · 1 tie
- AgentsCarrying out multi-step tasks on its own02Qwen3.5 35B A3B2 of 2 tests
- Long documentsFinding answers in very long texts02Qwen3.5 35B A3B2 of 2 tests
- Following instructionsDoing exactly what it is asked02Qwen3.5 35B A3B2 of 2 tests
- FactsGetting facts right instead of making them up11Even1 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.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Where Qwen3.5 35B A3B pulls ahead
- Reasons across sets of long documentsAA-LCR+51points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+41points ahead
- Graduate-level biology, physics and chemistry questionsGPQA Diamond+38.9points ahead
Where Qwen3.5 2B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+11.2points ahead
Every test, side by side
All 36 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 35B A3B
- Terminal-Bench 2.1Qwen3.5 35B A3B by 37.8340.8+37.8
- SciCodeQwen3.5 35B A3B by 34.92.837.7+34.9
- Terminal-Bench HardQwen3.5 35B A3B by 22.73.826.5+22.7
AgentsQwen3.5 35B A3B
- GDPValQwen3.5 35B A3B by 5.405.4+5.4
- τ-Bench V3 · BankingQwen3.5 35B A3B by 2.42.54.9+2.4
ReasoningQwen3.5 35B A3B
- GPQA DiamondQwen3.5 35B A3B by 38.945.684.5+38.9
- Humanity's Last ExamQwen3.5 35B A3B by 18.42.621+18.4
- CritPttie00.9tie
FactsEven
- AA-Omniscience · AccuracyQwen3.5 35B A3B by 128.120.1+12
- AA-Omniscience · Non-hallucinationQwen3.5 2B by 11.225.814.6+11.2
Images and chartsQwen3.5 35B A3B
Long documentsQwen3.5 35B A3B
- AA-LCRQwen3.5 35B A3B by 512172+51
- LongBench v2Qwen3.5 35B A3B by 20.338.759+20.3
Following instructionsQwen3.5 35B A3B
- IFBenchQwen3.5 35B A3B by 4131.572.5+41
- Multi-ChallengeQwen3.5 35B A3B by 26.333.760+26.3
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.
- Artificial Analysis Coding IndexQwen3.5 35B A3B by 34.12.937+34.1
- t2-benchQwen3.5 35B A3B by 32.448.881.2+32.4
- WMT24++Qwen3.5 35B A3B by 30.545.876.3+30.5
- MMLU-ProXQwen3.5 35B A3B by 28.752.381+28.7
- SuperGPQAQwen3.5 35B A3B by 25.937.563.4+25.9
- IncludeQwen3.5 35B A3B by 24.355.479.7+24.3
- BFCLv4Qwen3.5 35B A3B by 23.743.667.3+23.7
- SimpleVQAQwen3.5 35B A3B by 23.135.258.3+23.1
- MMMLUQwen3.5 35B A3B by 22.163.185.2+22.1
- τ²-Bench Telecom (AA run)Qwen3.5 35B A3B by 20.26989.2+20.2
- RealWorldQAQwen3.5 35B A3B by 1965.184.1+19
- MMLU-ProQwen3.5 35B A3B by 18.866.585.3+18.8
- C-EvalQwen3.5 35B A3B by 1773.290.2+17
- MMLU-ReduxQwen3.5 35B A3B by 13.779.693.3+13.7
- IFEvalQwen3.5 35B A3B by 13.378.691.9+13.3
- AA IntelligenceQwen3.5 35B A3B by 12.46.919.3+12.4
- AA-OmniscienceQwen3.5 35B A3B by 12.1-60.2-48.1+12.1
- HallusionBenchQwen3.5 35B A3B by 2.465.567.9+2.4
Questions people ask
Which is better, Qwen3.5 2B or Qwen3.5 35B A3B?
Qwen3.5 35B A3B wins six of the seven areas where both have results: coding, agents, reasoning, images and charts, long documents and following instructions. Qwen3.5 2B wins none. They are level on facts.
Which is better for coding?
Qwen3.5 35B A3B. It wins 3 of the 3 coding tests both models report; Qwen3.5 2B wins none.
How do you compare the two?
We use the 36 benchmark tests both models have published scores on. The verdict counts the 18 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.