Inference
EN

Qwen3.5 397B vs Qwen3.6 Plus

Change models
Qwen3.5 397BAlibaba
Qwen3.6 PlusAlibaba
IndexAll categories, out of 100.
53.8 ±2.4
54.6 ±2.1 best
GapIndex points below the top.
−0.8 · no clear gap
Top
CoverageIndex weight that has results.
100%
100%
SpeedOutput tokens per second.
59/s best
44/s
EloLMArena rating and rank.
—
1436
ReleasedMonth of release.
2026-02
2026-04
Input / 1MUS dollars per 1M input tokens.
$0.60
$0.325 best
Output / 1MUS dollars per 1M output tokens.
$3.60
$1.95 best
1K chats2K tokens in, 500 out, each.
$3.00
$1.63 best
1K documents20K tokens in, 1K out, each.
$15.60
$8.45 best
1K agent tasks200K tokens in, 20K out, each.
$192.00
$104.00 best
ContextMaximum tokens in one request.
128K
1M best
AgenticMulti-step tasks with tools.
53.3
55.0 best
CodingCode writing and repair.
53.3
56.2 best
ReasoningLogic problems and puzzles.
51.9 best
51.2
MultimodalTasks with images and text.
55.6
56.1 best
KnowledgeFacts and expert knowledge.
53.8
54.1 best
MultilingualTasks in many languages.
55.6 best
54.1
InstructionTasks with strict prompt rules.
55.6 best
52.4
MathMath problems.
53.6
54.9 best

Benchmarks

47 shared
Only Qwen3.5 397B: 22Only Qwen3.6 Plus: 44
Qwen3.5 397B
Qwen3.6 Plus
Artificial Analysis Humanity's Last ExamKnowledge
29.0% best85 / 222
27.8%92 / 222
Artificial Analysis Long Context ReasoningReasoning
77.3%77 / 217
78.3% best63 / 217
Critical Physics TasksReasoning
1.7%109 / 214
2.9% best98 / 214
Artificial Analysis GPQA DiamondKnowledge
89.3% best45 / 199
88.2%52 / 199
τ²-Bench Tool-Agent-User EvaluationAgentic
95.6%18 / 149
97.7% best9 / 149
Artificial Analysis IFBenchInstruction
78.8% best6 / 146
75.2%25 / 146
Graduate-Level Google-Proof Q&AKnowledge
88.4%36 / 141
90.4% best27 / 141
Software Engineering Benchmark VerifiedCoding
76.2%42 / 114
78.8% best26 / 114
Artificial Analysis MMMU-ProMultimodal
77.3%40 / 113
78.0% best39 / 113
SWE-bench ProCoding
50.9%69 / 83
56.6% best47 / 83
Humanity's Last ExamKnowledge
28.7%52 / 77
28.8% best51 / 77
Massive Multitask Language Understanding ProfessionalKnowledge
87.8%5 / 75
88.5% best3 / 75
Instruction-Following EvalInstruction
92.6%16 / 70
94.3% best8 / 70
LiveCodeBench v6Coding
83.6%25 / 59
87.1% best17 / 59
Instruction Following BenchmarkInstruction
76.5% best25 / 57
75.8%28 / 57
Massive Multi-discipline Multimodal Understanding ProMultimodal
79.0% best19 / 53
78.8%21 / 53
Gert Labs Composite Game BenchmarkAgentic
46.8%27 / 52
50.6% best21 / 52
MCP AtlasAgentic
46.1%47 / 52
48.2% best45 / 52
CharXiv ReasoningMultimodal
80.8%31 / 49
81.5% best27 / 49
Claw-EvalAgentic
56.8%34 / 42
58.8% best29 / 42