Qwen3 7 Max Chinas Latest Top Model

2026-05-25 · Source: The Batch | DeepLearning.AI | AI News & Insights - www.deeplearning.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Intermediate, medium

Summary

Alibaba's Qwen3.7-Max, released on May 20, is a reasoning model built for multi-step agent tasks, featuring a one million token context window and an extended-thinking mode. While it shows increased latency and reduced accuracy on simple factual questions (down 7.6 percentage points on AA-Omniscience), it excels in scientific reasoning and agentic tasks, demonstrating over 1,000 tool calls and code modifications to optimize a chip kernel, claiming a 10x speedup. It ranks fifth on the Artificial Analysis Intelligence Index, behind GPT-5.5 and Claude Opus 4.7. Concurrently, Microsoft Research released Fara1.5, a family of computer-use agents (4B, 9B, 27B) that operate browsers, achieving 72 percent on Online-Mind2Web, outperforming OpenAI's Operator (58.3 percent) and Google's Gemini 2.5 Computer Use (57.3 percent). Its FaraGen1.5 synthetic data pipeline enables training on gated applications. Nvidia also introduced Gated DeltaNet-2, an attention alternative with 1.3 billion parameters, outperforming Mamba-3 on long-context benchmarks by decoupling linear attention constraints. Cohere's Command A+, a 218 billion parameter mixture-of-experts model, offers sovereign AI capabilities, running on two H100 GPUs with 4-bit quantization and showing significant performance gains in agentic coding and telecom tasks. Meanwhile, U.S. President Donald Trump postponed an executive order regulating AI models, citing concerns about weakening America's technological edge.

Key takeaway

For ML engineers evaluating new model architectures or deploying agentic systems, recent advancements offer significant performance and efficiency gains. Consider Nvidia's Gated DeltaNet-2 for long-context tasks, or Cohere's Command A+ for sovereign AI deployments on limited hardware. Microsoft's Fara1.5 agents provide robust browser automation, while Alibaba's Qwen3.7-Max excels in complex multi-step reasoning, despite potential latency tradeoffs. Stay informed on regulatory shifts, as they directly impact deployment strategies.

Key insights

AI models are demonstrating advanced reasoning and agentic capabilities, alongside architectural innovations and evolving regulatory landscapes.

Principles

• Extended context improves agentic task performance.
• Decoupling attention gates enhances long-context processing.
• Synthetic data pipelines enable training on gated applications.

Method

Microsoft's FaraGen1.5 pipeline creates functional clones of gated applications for training computer-use agents. Nvidia's Gated DeltaNet-2 splits key/value gates for linear attention.

In practice

• Deploy sovereign AI with Cohere's Command A+ on two H100 GPUs.
• Utilize Qwen3.7-Max for complex multi-step agent tasks.
• Implement browser automation with Microsoft's Fara1.5 agents.

Topics

• Agentic AI
• Large Language Models
• Attention Mechanisms
• Sovereign AI
• AI Regulation
• Computer Use Agents

Code references

• NVlabs/GatedDeltaNet-2

Best for: AI Engineer, NLP Engineer, CTO, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Qwen3 7 Max Adds Speed And Power — Alibaba's Qwen3.7-Max demonstrates high performance and speed, particularly for agentic tasks, despite its closed-source nature.
• Qwen 3.7 Max: NEW Powerful AI Model! Beats Opus 4.6, Gemini 3.1, Deepseek v4! (Fully Tested) — Qwen 3.7 Max excels in long-horizon autonomous coding and complex task execution, rivaling frontier models in efficiency and performance.
• Introducing Command A+: Making sovereign agentic capabilities available to all - Cohere — Cohere's Command A+ democratizes enterprise-grade agentic AI through an efficient, open-source MoE model.
• QWEN 3.7 MAX - Thinking: A Surprise 😄 — Qwen 3.7 Max shows strong complex problem-solving, outperforming many peers on benchmarks and intricate live tests.
• 🇨🇳 Alibaba unveils new Qwen3.5 model for ‘agentic AI era’, Qwen3.5-397B-A17B. Apache 2.0 license — Sparse Mixture-of-Experts and hybrid attention enhance LLM efficiency and long-context performance for agentic AI.
• Alibaba's proprietary Qwen3.7-Max can run for 35 hours autonomously and supports external harnesses like Anthropic's Claude Code — Autonomous AI agents are now capable of sustained, complex task execution over multiple days, marking a new era.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Batch | DeepLearning.AI | AI News & Insights - www.deeplearning.ai.