Qwen3.5 397B A17B vs Qwen3.5 9B
Wins 6 of 8 areas
Coding · Agents · Reasoning · Images and charts · Long documents · Following instructions
Wins 0 of 8 areas
—
Qwen3.5 397B A17B is the stronger all-rounder.Qwen3.5 9B is cheaper.
Scores updated · 41 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 software40Qwen3.5 397B A17B4 of 4 tests
- ReasoningHard problems that need careful thinking30Qwen3.5 397B A17B3 of 3 tests
- AgentsCarrying out multi-step tasks on its own20Qwen3.5 397B A17B2 of 3 tests · 1 tie
- 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
- Images and chartsUnderstanding pictures, charts and video10Qwen3.5 397B A17B1 of 1 test
- FactsGetting facts right instead of making them up11Even1 each
- MathCompetition and research-level math11Even1 each
Images and charts 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 9B costs 92% 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
- Command-line tasks in a real terminalTerminal-Bench 2.1+22.1points ahead
- Recent programming contest problemsLiveCodeBench v6+18points ahead
- Hard command-line tasks in a real terminalTerminal-Bench Hard+16.7points ahead
Where Qwen3.5 9B pulls ahead
- Avoids making up answers it doesn't knowAA-Omniscience · Non-hallucination+5.3points ahead
- US invitational high-school math exam problemsAIME 2026+1.2points ahead
Every test, side by side
All 41 tests both models report. The winning score is in its model's colour; marks a score checked independently.
CodingQwen3.5 397B A17B
- Terminal-Bench 2.1Qwen3.5 397B A17B by 22.151.329.2+22.1
- LiveCodeBench v6Qwen3.5 397B A17B by 1883.665.6+18
- Terminal-Bench HardQwen3.5 397B A17B by 16.740.924.2+16.7
- SciCodeQwen3.5 397B A17B by 15.344.829.5+15.3
AgentsQwen3.5 397B A17B
- GDPValQwen3.5 397B A17B by 14.814.80+14.8
- τ-Bench V3 · BankingQwen3.5 397B A17B by 6.413.47+6.4
- Terminal-Bench 4.0tie00.5tie
ReasoningQwen3.5 397B A17B
- Humanity's Last ExamQwen3.5 397B A17B by 14.12914.9+14.1
- GPQA DiamondQwen3.5 397B A17B by 8.789.380.6+8.7
- CritPtQwen3.5 397B A17B by 1.41.70.3+1.4
FactsEven
- AA-Omniscience · AccuracyQwen3.5 397B A17B by 14.430.816.4+14.4
- AA-Omniscience · Non-hallucinationQwen3.5 9B by 5.311.116.4+5.3
Images and chartsQwen3.5 397B A17B
- MMMU-ProQwen3.5 397B A17B by 8.177.369.2+8.1
MathEven
- HMMT Feb. 2026Qwen3.5 397B A17B by 16.787.971.2+16.7
- AIME 2026Qwen3.5 9B by 1.291.392.5+1.2
Long documentsQwen3.5 397B A17B
- LongBench v2Qwen3.5 397B A17B by 863.255.2+8
- AA-LCRQwen3.5 397B A17B by 7.377.370+7.3
Following instructionsQwen3.5 397B A17B
- Multi-ChallengeQwen3.5 397B A17B by 13.167.654.5+13.1
- IFBenchQwen3.5 397B A17B by 12.178.866.7+12.1
Other results22 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.
- AA-OmniscienceQwen3.5 397B A17B by 22.8-30.8-53.5+22.8
- VITA-BenchQwen3.5 397B A17B by 19.949.729.8+19.9
- Artificial Analysis Coding IndexQwen3.5 397B A17B by 19.548.228.7+19.5
- DeepPlanningQwen3.5 397B A17B by 16.334.318+16.3
- PinchBenchQwen3.5 397B A17B by 15.186.671.5+15.1
- Toolathlon VerifiedQwen3.5 397B A17B by 14.840.725.9+14.8
- BFCLv4Qwen3.5 397B A17B by 12.872.960.1+12.8
- SuperGPQAQwen3.5 397B A17B by 12.270.458.2+12.2
- IncludeQwen3.5 397B A17B by 1085.675.6+10
- HMMT 2025Qwen3.5 397B A17B by 9.892.782.9+9.8
- τ²-Bench Telecom (AA run)Qwen3.5 397B A17B by 8.895.686.8+8.8
- MMLU-ProXQwen3.5 397B A17B by 8.484.776.3+8.4
- τ³-Bench BankingQwen3.5 397B A17B by 8.213.45.2+8.2
- t2-benchQwen3.5 397B A17B by 7.686.779.1+7.6
- MMMLUQwen3.5 397B A17B by 7.388.581.2+7.3
- AA IntelligenceQwen3.5 397B A17B by 7.218.411.2+7.2
- MMLU-ProQwen3.5 397B A17B by 5.387.882.5+5.3
- C-EvalQwen3.5 397B A17B by 4.89388.2+4.8
- Claw-Eval AvgQwen3.5 397B A17B by 4.270.766.5+4.2
- MMLU-ReduxQwen3.5 397B A17B by 3.894.991.1+3.8
- AA Agentic IndexQwen3.5 397B A17B by 3.610.67+3.6
- IFEvalQwen3.5 397B A17B by 1.192.691.5+1.1
Questions people ask
Which is better, Qwen3.5 397B A17B or Qwen3.5 9B?
Qwen3.5 397B A17B wins six of the eight areas we test: coding, agents, reasoning, images and charts, long documents and following instructions. Qwen3.5 9B wins none, but costs 92% less. They are level on facts and math.
Which is better for coding?
Qwen3.5 397B A17B. It wins 4 of the 4 coding tests both models report; Qwen3.5 9B 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 9B costs $0.14 and $0.20. That makes Qwen3.5 9B about 92% cheaper for the same work.
How do you compare the two?
We use the 41 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 22 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.