🕷️
SetupAI
🔥Stalecollected in 8h

Alibaba's Qwen3.7-Max Ranks Top 5 Globally

Alibaba's Qwen3.7-Max Ranks Top 5 Globally
PostLinkedIn
🔥Read original on 36氪
#llm-benchmark#chinese-ai#api-serviceqwen3.7-maxalibabaqwen3.7-maxartificialanalysisalibaba-cloud

💡New Chinese flagship model Qwen3.7-Max hits top 5 global ranking, challenging GPT and Claude performance.

⚡ 30-Second TL;DR

What Changed

Qwen3.7-Max achieved a score of 56.6, ranking 5th globally and 1st among Chinese models.

Why It Matters

The rise of Qwen3.7-Max provides developers with a high-performance alternative to Western models, particularly for applications requiring strong Chinese language capabilities and domestic infrastructure compliance.

What To Do Next

Integrate the Qwen3.7-Max API via Alibaba Cloud Bailian to benchmark its performance against your current LLM stack for Chinese-centric tasks.

Who should care:Developers & AI Engineers

Key Points

  • •Qwen3.7-Max achieved a score of 56.6, ranking 5th globally and 1st among Chinese models.
  • •Performance is reported to be competitive with top-tier models like GPT, Claude, and Gemini.
  • •The model will be accessible for developers via the Alibaba Cloud Bailian API service.

🧠 Deep Insight

Web-grounded analysis with 15 cited sources.

🔑 Enhanced Key Takeaways

  • •Alibaba's Qwen3.7-Max achieved a 4.8-point increase over its predecessor, Qwen3.6 Max Preview (51.8), and surpassed Google's Gemini 3.5 Flash (55.3) on the ArtificialAnalysis Intelligence Index.
  • •The model's performance improvements are primarily concentrated in scientific reasoning, agentic capability, and coding, with significant gains observed in CritPt, Humanity's Last Exam, and Terminal-Bench Hard scores.
  • •Qwen3.7-Max features a 1M token context window, a substantial increase from the 256K tokens offered by Qwen3.6 Max Preview, and is currently limited to text input and output.
  • •A portion of Qwen3.7-Max's elevated ArtificialAnalysis score is attributed to a strategy of increased abstention on the AA-Omniscience benchmark, which led to a decrease in its raw accuracy but a notable reduction in its hallucination rate.
  • •Qwen3.7-Max is positioned as a versatile agent foundation, demonstrating strengths in writing and debugging code, automating office workflows, and executing autonomous tasks over extended sequences.
📊 Competitor Analysis▸ Show

Competitor Analysis: Qwen3.7-Max vs. Top-Tier Models

ModelArtificialAnalysis Intelligence Index Score (v4.0.4)Context Window (Tokens)Input Pricing (per 1M tokens)Output Pricing (per 1M tokens)
GPT-5.560.2N/AN/AN/A
Claude Opus 4.757.3N/AN/AN/A
Gemini 3.1 Pro Preview57.2N/AN/AN/A
Qwen3.7-Max56.61,000,000$2.50$7.50
Gemini 3.5 Flash55.3N/AN/AN/A

Note: Pricing and context window details for competitor models (GPT-5.5, Claude Opus 4.7, Gemini 3.1 Pro Preview, Gemini 3.5 Flash) were not explicitly available in the search results for these specific versions. Pricing for Qwen3.7-Max is based on its listed rates as of May 20, 2026.

🛠️ Technical Deep Dive

  • The Qwen model family generally employs a transformer-based architecture, incorporating advancements in attention mechanisms and training methodologies.
  • Qwen3 models integrate distinct "thinking" and "non-thinking" modes, allowing for flexible control over reasoning depth and response speed, and support 119 languages and dialects.
  • Some Qwen models, such as Qwen3.6-35B-A3B, utilize a hybrid sparse Mixture-of-Experts (MoE) architecture, which combines Gated DeltaNet linear attention with standard gated attention layers to achieve efficient inference.
  • Qwen-Image, a related image generation model within the Qwen series, features a three-core module architecture comprising a Multimodal Large Language Model (e.g., Qwen2.5-VL), a Variational AutoEncoder (VAE), and a Multimodal Diffusion Transformer (MMDiT) that includes Multimodal Scalable RoPE (MSRoPE) for joint text-image positional encoding.
  • Qwen3.7-Max supports explicit prompt caching, which contributes to more efficient use of repeated context.

🔮 Future ImplicationsAI analysis grounded in cited sources

Alibaba will strengthen its position in the global AI agent market.
Qwen3.7-Max's explicit design as a versatile agent foundation, with strengths in coding, office automation, and autonomous execution, indicates a strategic focus on the growing AI agent sector.
The Qwen series will continue to drive the adoption of hybrid open-source and proprietary AI models.
Alibaba consistently releases both open-source and proprietary Qwen models, fostering a broad developer ecosystem while also offering advanced commercial solutions, which promotes a hybrid adoption strategy.
Alibaba Cloud's Bailian platform will become a more central hub for diverse AI model access, particularly in China.
The full opening of the Bailian platform and its partnerships with other prominent AI companies like Zhipu AI and Minimax indicate its strategic importance as a comprehensive model service and API management platform.

⏳ Timeline

2023-04
Alibaba launched a beta of Qwen (Tongyi Qianwen).
2023-08
Alibaba released its first open-model, Qwen-7B.
2023-09
Qwen opened for public use after regulatory clearance.
2024-06
Qwen2 series released, including open-weight models.
2025-04-29
Qwen3 series released, expanding multilingual support to 119 languages.
2026-05-18
Qwen3.7 Max and Qwen3.7 Plus stable releases.
🔥Read original article on 36氪
📰

Weekly AI Recap

Read this week's curated digest of top AI events →

👉Related Updates

Same topic

Explore #llm-benchmark

Same product

More on qwen3.7-max

Same source

Latest from 36氪

AI-curated news aggregator. All content rights belong to original publishers.
Original source: 36氪 ↗