Model

Qwen3.7-Max

CLOSED

Alibaba · Qwen3.7 family · released May 20, 2026

Agent-first flagship with 1M-token context (up from 256K on the prior generation); Alibaba reports 1,000+ consecutive tool calls and a 35-hour autonomous kernel-optimization run in internal testing (not independently verified). Not open-weight; no HF checkpoints as of this research.

ReasoningCodingVisionFunction callingTool useAgentic

2564.6

Elo · rank #12

Parameters

Undisclosed

Active params

Undisclosed

Context

1M tokens

Architecture

MoE, exact expert configuration undisclosed; positioned as an agent-first reasoning model

License

Proprietary

Languages

—

API price (in/out)

$2.5 / $7.5

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

Agents' Last ExamAgents31.1%#12

best: GPT-5.6 · 52.7%

GDPval-AAAgents1280#23

best: Claude Fable 5 · 1932

GPQA DiamondReasoning92.4%#15

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning41.4%#22

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning79.1%#8

best: MiniMax M3 · 83.0%

IFEvalReasoning94.3%#6

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding91.6%#3

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MMLU-ProKnowledge89.6%#3

best: Claude Fable 5 · 91.5%

MMLU-ReduxKnowledge95.0%#1

best: this model · 95.0%

MMMLUKnowledge90.3%#6

best: Gemini 3.1 Pro · 92.6%

SuperGPQAReasoning73.6%#1

best: this model · 73.6%

SWE-bench ProCoding60.6%#9

best: Claude Fable 5 · 80.0%

SWE-bench VerifiedCoding80.4%#16

best: Claude Fable 5 · 95.0%

Terminal-Bench 2.0Coding69.7%#16

best: GPT-5.6 · 88.8%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$2.5

Output / M tok

$7.5

Qwen3.7 family

Elo progression across releases

Qwen3.7-MaxMay 20262564.6 Qwen3.7-PlusJun 20262436.9

Resources

DOCSwww.alibabacloud.com/help/en/model-studio/models APIAvailable via Alibaba API

API price $2.5/$7.5 · each benchmark row carries its own source badge (see methodology)