Is Qwen3.5 397B A17B better than Qwen3.6 35B A3B?

On accuracy they are statistically tied: Qwen3.5 397B A17B scored 32.5% and Qwen3.6 35B A3B 37.5% on the same 80 questions from open datasets. Similar accuracy; Qwen3.6 35B A3B costs less per question.

Which is cheaper, Qwen3.5 397B A17B or Qwen3.6 35B A3B?

Qwen3.6 35B A3B is cheaper: $0.0037 per question as billed, against $0.013. List prices: Qwen3.5 397B A17B $0.5 in / $3.6 out, Qwen3.6 35B A3B $0.1 in / $0.95 out per million tokens.

Which is faster, Qwen3.5 397B A17B or Qwen3.6 35B A3B?

Qwen3.5 397B A17B is faster at 240 output tokens per second, against 135. Time to first token: 0.6 s for Qwen3.5 397B A17B, 0.4 s for Qwen3.6 35B A3B.

Which has the larger context window, Qwen3.5 397B A17B or Qwen3.6 35B A3B?

Both accept about the same amount of context: 262k and 262k tokens.

Head to head · tested 27 Oct 2026

Qwen3.5 397B A17B vs Qwen3.6 35B A3B: accuracy, speed and price

Similar accuracy; Qwen3.6 35B A3B costs less per question.

QwenQwen3.5 397B A17B32.5% score · #62$0.013 per question240 tok/s

QwenQwen3.6 35B A3B37.5% score · #59$0.0037 per question135 tok/s

Verdict

Built from the measurements below

Accuracy is too close to call: Qwen3.5 397B A17B scored 32.5% and Qwen3.6 35B A3B 37.5% on the same 80 questions, a gap within what chance alone could produce.
Qwen3.6 35B A3B is 3.6× cheaper per question as billed ($0.0037 vs $0.013), reasoning tokens included.
Qwen3.5 397B A17B generates text 1.8× faster: 240 tokens per second against 135.
Qwen3.6 35B A3B starts answering sooner: 0.4 s to the first token, against 0.6 s for Qwen3.5 397B A17B.

Which should you choose?

Choose Qwen3.5 397B A17B for faster streaming. Choose Qwen3.6 35B A3B for the lower cost per correct answer ($0.0099 against $0.041) and a quicker first token. The widest subject gap is in medicine: Qwen3.6 35B A3B answered 6 of 11 correctly, Qwen3.5 397B A17B 2 of 11. With so few questions per subject, treat this as a lead to test on your own prompts, not a result.

Accuracy gap -5.0 points (z = -0.66). A winner is called only when z is at least 1.96, the usual 95% threshold. Speed, latency and price need a 10% margin.

Head to head

Same questions, same prompt, same API

Measure	Qwen3.5 397B A17B	Qwen3.6 35B A3B	Edge
Verification scoreHigher is better	32.5%	37.5%	Too close to call
Cost per questionLower is better	$0.013	$0.0037	Qwen3.6 35B A3B
Output speedHigher is better	240	135	Qwen3.5 397B A17B
Time to first tokenLower is better	0.6 s	0.4 s	Qwen3.6 35B A3B
End-to-end response timeLower is better	3.5 s	5.6 s	Qwen3.5 397B A17B
Tokens per questionLower is better	4,010	3,563	Qwen3.6 35B A3B
Context windowHigher is better	262k	262k	Similar

Side by side

Bars are scaled to the larger of the two values. The winner on each measure is marked in bold.

Every measure

Qwen3.5 397B A17B first, Qwen3.6 35B A3B second

Verification scorehigher is better

Qwen3.5 397B A17B32.5%

Qwen3.6 35B A3B37.5%

Cost per questionlower is better

Qwen3.5 397B A17B$0.013

Qwen3.6 35B A3B$0.0037

Output speedhigher is better

Qwen3.5 397B A17B240

Qwen3.6 35B A3B135

Time to first tokenlower is better

Qwen3.5 397B A17B0.6 s

Qwen3.6 35B A3B0.4 s

End-to-end response timelower is better

Qwen3.5 397B A17B3.5 s

Qwen3.6 35B A3B5.6 s

Tokens per questionlower is better

Qwen3.5 397B A17B4,010

Qwen3.6 35B A3B3,563

Context windowhigher is better

Qwen3.5 397B A17B262k

Qwen3.6 35B A3B262k

Score vs. cost per question

Both models in color, every other model in grey

Results by subject and dataset

Questions answered correctly. Small samples: a gap of one or two questions is noise.

Subject	Questions	Qwen3.5 397B A17B	Qwen3.6 35B A3B
Science	30	10/30	9/30
Engineering	15	0/15	2/15
Medicine	11	2/11	6/11
Economics	9	5/9	5/9
Law	6	1/6	1/6
Humanities & social science	9	8/9	7/9
SuperGPQA dataset	50	5/50	9/50
MMLU-Pro dataset	30	21/30	21/30

Specifications and pricing

From OpenRouter's catalogue, plus the leaderboard's test dates

Specification	Qwen3.5 397B A17B	Qwen3.6 35B A3B
Creator	Qwen	Qwen
Released	16 Feb 2026	27 Apr 2026
Weights	Open weights	Open weights
Reasoning	Yes	Yes
Input	text, image, video	text, image, video
Output	text	text
Context window	262k tokens	262k tokens
Knowledge cutoff	—	—
List price in / out per 1M	$0.5 / $3.6	$0.1 / $0.95
Blended price per 1M (3:1)	$1.27	$0.31
Cheapest host	Alibaba · $0.39 / $2.34	Darkbloom · $0.05 / $0.7
Hosts on OpenRouter	10	10
Last tested	27 Oct 2026	26 Oct 2026

Compare any two models →

Qwen3.5 397B A17B vs Qwen3.6 35B A3B: frequently asked questions

Is Qwen3.5 397B A17B better than Qwen3.6 35B A3B?: On accuracy they are statistically tied: Qwen3.5 397B A17B scored 32.5% and Qwen3.6 35B A3B 37.5% on the same 80 questions from open datasets. Similar accuracy; Qwen3.6 35B A3B costs less per question.
Which is cheaper, Qwen3.5 397B A17B or Qwen3.6 35B A3B?: Qwen3.6 35B A3B is cheaper: $0.0037 per question as billed, against $0.013. List prices: Qwen3.5 397B A17B $0.5 in / $3.6 out, Qwen3.6 35B A3B $0.1 in / $0.95 out per million tokens.
Which is faster, Qwen3.5 397B A17B or Qwen3.6 35B A3B?: Qwen3.5 397B A17B is faster at 240 output tokens per second, against 135. Time to first token: 0.6 s for Qwen3.5 397B A17B, 0.4 s for Qwen3.6 35B A3B.
Which has the larger context window, Qwen3.5 397B A17B or Qwen3.6 35B A3B?: Both accept about the same amount of context: 262k and 262k tokens.