Atlas

Models

← All models

Qwen3.6-Max-Preview

Alibaba · 2026-04-18

Qwen3.6-Max-Preview is an early preview of Alibaba's next proprietary Qwen model, with improvements in agentic coding, world knowledge, instruction following, and real-world agent reliability over Qwen3.6-Plus. It is a hosted text model available through Alibaba Cloud Model Studio with both thinking and non-thinking modes.

Benchmark scores

BenchmarkScore
AA-LCR69.7
AA-Omniscience Index10.2
Artificial Analysis Intelligence Index40.0
Artificial Analysis Omniscience Accuracy37.7
Artificial Analysis Omniscience Hallucination Rate44.2
BuseyBench SVG3.3
Buyout Game Bradley-Terry Rating1459
Capability151.5
CaseLaw47.9
Chess Puzzles17.0
CorpFin v266.5
CorpFin v2 - Exact Pages63.3
CorpFin v2 - Max Fitting Context68.2
CorpFin v2 - Shared Max Context67.9
CritPt (avg@5)3.7
Epoch Capabilities Index (ECI)149
Extended NYT Connections 940-Puzzle Set82.2
Extended NYT Connections Newest 10078.5
GBENCH One-shot Coding GScore42.1
GPQA Diamond89.1
GPQA Diamond (Artificial Analysis avg@5)88.8
Humanity's Last Exam (Text-Only)28.9
IFBench (Artificial Analysis)76.6
Judgemark v445.1
LLM Creative Story Writing-0.3
LLM Creative Story Writing Win Probability0.5
LLM Debate Benchmark1511
LLM Sycophancy Conditional Rate5.6
LLM Sycophancy Conditional Total Inconsistency26.4
LLM Sycophancy Contrarian Rate15.1
Loading Atlas data…