Qwen3.5 397B A17B vs Qwen3.5 4B
Wins 7 of 8 areas
Coding · Agents · Reasoning · Images and charts · Math · Long documents · Following instructions
Wins 0 of 8 areas
—
Qwen3.5 397B A17B is the stronger all-rounder.Qwen3.5 4B is cheaper.
Scores updated · 42 tests both models report · How we compare
Where each one wins
Tests won in each of the eight areas we test. 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 software50Qwen3.5 397B A17B5 of 5 tests
- ReasoningHard problems that need careful thinking30Qwen3.5 397B A17B3 of 3 tests
- Images and chartsUnderstanding pictures, charts and video20Qwen3.5 397B A17B2 of 2 tests
- MathCompetition and research-level math20Qwen3.5 397B A17B2 of 2 tests
- Long documentsFinding answers in very long texts20Qwen3.5 397B A17B2 of 2 tests
- Following instructionsDoing exactly what it is asked20Qwen3.5 397B A17B2 of 2 tests
- AgentsCarrying out multi-step tasks on its own10Qwen3.5 397B A17B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
Agents rests on a single test.
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.
Qwen3.5 4B costs 96% 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 Qwen3.5 397B A17B pulls ahead
- Code for real scientific research problemsSciCode+28.7points ahead
- Recent programming contest problemsLiveCodeBench v6+27.8points ahead
- Follows unfamiliar, precisely checkable instructionsIFBench+26.8points ahead
Where Qwen3.5 4B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+2.3points ahead
Every test, side by side
All 42 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 397B A17B
- SciCodeQwen3.5 397B A17B by 28.744.816.1+28.7
- LiveCodeBench v6Qwen3.5 397B A17B by 27.883.655.8+27.8
- Terminal-Bench 2.1Qwen3.5 397B A17B by 25.551.325.8+25.5
- Terminal-Bench HardQwen3.5 397B A17B by 22.740.918.2+22.7
- SWE-bench VerifiedQwen3.5 397B A17B by 16.976.459.5+16.9
ReasoningQwen3.5 397B A17B
- Humanity's Last ExamQwen3.5 397B A17B by 19.1299.9+19.1
- GPQA DiamondQwen3.5 397B A17B by 12.289.377.1+12.2
- CritPtQwen3.5 397B A17B by 1.71.70+1.7
FactsEven
- AA-Omniscience · AccuracyQwen3.5 397B A17B by 15.730.815.1+15.7
- AA-Omniscience · Non-hallucinationQwen3.5 4B by 2.311.113.4+2.3
Images and chartsQwen3.5 397B A17B
MathQwen3.5 397B A17B
- HMMT Feb. 2026Qwen3.5 397B A17B by 1587.972.9+15
- AIME 2026Qwen3.5 397B A17B by 1.691.389.7+1.6
Long documentsQwen3.5 397B A17B
- AA-LCRQwen3.5 397B A17B by 14.377.363+14.3
- LongBench v2Qwen3.5 397B A17B by 13.263.250+13.2
Following instructionsQwen3.5 397B A17B
- IFBenchQwen3.5 397B A17B by 26.878.852+26.8
- Multi-ChallengeQwen3.5 397B A17B by 18.667.649+18.6
Other results23 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.
- HLE (with tools)Qwen3.5 397B A17B by 36.748.311.6+36.7
- VITA-BenchQwen3.5 397B A17B by 27.749.722+27.7
- AA-OmniscienceQwen3.5 397B A17B by 27.6-30.8-58.4+27.6
- Artificial Analysis Coding IndexQwen3.5 397B A17B by 25.648.222.6+25.6
- BFCLv4Qwen3.5 397B A17B by 22.372.950.6+22.3
- AA Agentic IndexQwen3.5 4B by 21.910.632.5+21.9
- SuperGPQAQwen3.5 397B A17B by 17.570.452.9+17.5
- DeepPlanningQwen3.5 397B A17B by 16.734.317.6+16.7
- HMMT 2025Qwen3.5 397B A17B by 15.992.776.8+15.9
- PinchBenchQwen3.5 397B A17B by 15.386.671.3+15.3
- IncludeQwen3.5 397B A17B by 14.685.671+14.6
- MMLU-ProXQwen3.5 397B A17B by 13.284.771.5+13.2
- MMMLUQwen3.5 397B A17B by 12.488.576.1+12.4
- WMT24++Qwen3.5 397B A17B by 12.378.966.6+12.3
- MMLU-ProQwen3.5 397B A17B by 8.787.879.1+8.7
- Claw-Eval AvgQwen3.5 397B A17B by 8.470.762.3+8.4
- C-EvalQwen3.5 397B A17B by 7.99385.1+7.9
- τ³-Bench BankingQwen3.5 397B A17B by 7.913.45.5+7.9
- t2-benchQwen3.5 397B A17B by 6.886.779.9+6.8
- MMLU-ReduxQwen3.5 397B A17B by 6.194.988.8+6.1
- AA IntelligenceQwen3.5 397B A17B by 5.318.413.1+5.3
- τ²-Bench Telecom (AA run)Qwen3.5 397B A17B by 3.595.692.1+3.5
- IFEvalQwen3.5 397B A17B by 2.892.689.8+2.8
Questions people ask
Which is better, Qwen3.5 397B A17B or Qwen3.5 4B?
Qwen3.5 397B A17B wins seven of the eight areas we test: coding, agents, reasoning, images and charts, math, long documents and following instructions. Qwen3.5 4B wins none, but costs 96% less. They are level on facts.
Which is better for coding?
Qwen3.5 397B A17B. It wins 5 of the 5 coding tests both models report; Qwen3.5 4B wins none.
Which is cheaper?
Qwen3.5 397B A17B costs $0.60 per million input tokens and $3.60 per million output tokens; Qwen3.5 4B costs $0.03 and $0.15. That makes Qwen3.5 4B about 96% cheaper for the same work.
How do you compare the two?
We use the 42 benchmark tests both models have published scores on. The verdict counts the 19 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 23 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.