Inference
EN

Qwen3.6-35B-A3B vs Qwen 3.6 Max (preview)

Change models
Qwen3.6-35B-A3BAlibaba
Qwen 3.6 Max (preview)Alibaba
IndexAll categories, out of 100.
50.6 ±3.1
57.2 ±5.0 best
GapIndex points below the top.
−6.6
Top
CoverageIndex weight that has results.
95% best
85%
SpeedOutput tokens per second.
65/s
75/s best
EloLMArena rating and rank.
—
1446
ReleasedMonth of release.
2026-04
2026-04
Input / 1MUS dollars per 1M input tokens.
$0.10 best
$1.03
Output / 1MUS dollars per 1M output tokens.
$0.90 best
$6.16
1K chats2K tokens in, 500 out, each.
$0.65 best
$5.14
1K documents20K tokens in, 1K out, each.
$2.90 best
$26.70
1K agent tasks200K tokens in, 20K out, each.
$38.00 best
$328.64
ContextMaximum tokens in one request.
262K
262K
AgenticMulti-step tasks with tools.
52.7
59.2 best
CodingCode writing and repair.
49.6
57.0 best
ReasoningLogic problems and puzzles.
49.9
53.4 best
MultimodalTasks with images and text.
51.5
—
KnowledgeFacts and expert knowledge.
49.8
59.3 best
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
49.2
60.6 best
MathMath problems.
49.0
53.2 best

Benchmarks

14 shared
Only Qwen3.6-35B-A3B: 39Only Qwen 3.6 Max (preview): 14
Qwen3.6-35B-A3B
Qwen 3.6 Max (preview)
Artificial Analysis Humanity's Last ExamKnowledge
22.2%106 / 222
30.8% best73 / 222
Artificial Analysis Long Context ReasoningReasoning
71.7%106 / 217
80.7% best37 / 217
GPQA diamond · Epoch AIKnowledge
84.8%72 / 217
87.4% best63 / 217
Critical Physics TasksReasoning
0.3%137 / 214
3.7% best93 / 214
OTIS Mock AIME 2024-2025Math
68.9%102 / 200
91.1% best58 / 200
Artificial Analysis GPQA DiamondKnowledge
84.1%85 / 199
88.8% best49 / 199
τ²-Bench Tool-Agent-User EvaluationAgentic
95.3%21 / 149
95.9% best13 / 149
Artificial Analysis IFBenchInstruction
64.4%57 / 146
76.6% best11 / 146
SWE-bench ProCoding
49.5%72 / 83
57.3% best43 / 83
Mystery Game PuzzlesReasoning
22.0% best41 / 80
19.0%46 / 80
Terminal-Bench 2.0 · BenchLMAgentic
51.5%27 / 41
65.4% best12 / 41
NL2RepoCoding
29.4%35 / 36
42.9% best22 / 36
SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesKnowledge
64.7%17 / 22
73.9% best3 / 22
QwenClawBenchAgentic
52.6%8 / 10
59.0% best3 / 10