來源SCMP Technology•較早收集於 1m
阿里巴巴 Qwen3.5-Max 領先中國，落後美國
#china-ai#benchmarks#model-previewqwen3.5-max-previewalibabaqwen3.5-max-previewarenaanthropicopenai
💡中國頂尖 LLM 基準勝對手，縮小與美國差距（22字）
⚡ 30 秒速覽
有什麼變化
Qwen3.5-Max-Preview 在 Arena 排行超越中國 AI 模型
為什麼重要
強化中國本土 AI 生態，為從業人員提供高性能美國模型替代品。可能加速多模態 LLM 競爭與創新。
下一步行動
在 Arena 上測試 Qwen3.5-Max-Preview，對比 Claude 和 GPT 模型基準。
誰應關注：Researchers & Academics
關鍵要點
- •Qwen3.5-Max-Preview 在 Arena 排行超越中國 AI 模型
- •落後 Anthropic、Google、OpenAI 等美國領先者
- •阿里巴巴 Qwen 3.5 系列旗艦模型現開放預覽
- •確立阿里巴巴中國 AI 龍頭地位
🧠 深度解析
背景與延伸：來自公開資料，非原文內容。引用 7 個來源。
🔑 增強重點摘要
- •The Qwen 3.5 family introduces a novel 'Hybrid Mixture-of-Experts' architecture utilizing Gated DeltaNet (linear attention), which enables a 1-million token context window while delivering up to 19x higher decoding throughput than the previous Qwen 3 generation.
- •Alibaba has expanded linguistic support to 201 languages and dialects, utilizing a massive 250,000-token vocabulary that improves encoding efficiency by up to 60% for non-English scripts compared to the 150,000-token limit in Qwen 3.
- •The release follows a significant leadership exodus in early 2026, including the departure of technical lead Lin Junyang (Justin Lin) and head of post-training Yu Bowen, sparking industry debate over Alibaba's long-term commitment to its open-source strategy.
- •Qwen 3.5-Max-Preview features a dual-mode 'Thinking' vs. 'Fast' inference capability, where the model can engage in internal chain-of-thought reasoning (via
tags) to match US rivals in complex logic while maintaining a low-latency mode for routine tasks.
📊 競品分析▸ Show
| Feature | Qwen 3.5-Max-Preview | Gemini 3.1 Pro | Claude 4.6 Opus | GPT-5.4 |
|---|---|---|---|---|
| Arena Elo | ~1451 | 1505 | 1503 | 1485 |
| Context Window | 1M (Hosted) / 262K (Native) | 2M+ | 200K | 128K |
| Architecture | 397B MoE (17B Active) | Proprietary MoE | Proprietary | Proprietary |
| License | Apache 2.0 (Open-Weight) | Proprietary | Proprietary | Proprietary |
| Multilingual | 201 Languages | 150+ Languages | 95+ Languages | 100+ Languages |
| Pricing (per 1M) | ~$0.10 (Est. API) | $1.25 (Input) | $3.00 (Input) | $2.50 (Input) |
🛠️ 技術深入
Detailed technical specifications for the Qwen 3.5-397B-A17B model:
- Parameter Count: 397 billion total parameters with a sparse Mixture-of-Experts (MoE) routing that activates only 17 billion parameters per token.
- Attention Mechanism: A hybrid layout consisting of 60 layers where 15 groups of 3 'Gated DeltaNet' (linear attention) layers are interleaved with 1 'Gated Attention' layer to optimize memory usage for long-context sequences.
- Multimodal Integration: Native 'early-fusion' vision-language architecture where text and visual tokens are processed within the same transformer backbone rather than using a separate adapter.
- Training Scale: Pre-trained on an estimated 36+ trillion tokens with a heavy emphasis on synthetic 'agentic' data and reinforcement learning (RL) scaled across million-agent environments.
- Inference Optimizations: Native support for Multi-Token Prediction (MTP) and SGLang/vLLM acceleration engines, achieving near-100% multimodal training efficiency.
🔮 前景展望基於引用來源的 AI 分析
Alibaba will pivot toward a 'Cloud-First' proprietary model tier.
The increasing performance gap between the open-weight 397B model and the closed-source 'Plus' and 'Max-Preview' versions suggests Alibaba is prioritizing its Model Studio ecosystem over pure open-source parity.
Qwen will become the dominant foundation for non-English AI agents.
With support for 201 languages and superior performance on regional benchmarks like C-Eval, it is positioned as the primary alternative to US models in Southeast Asia and the Middle East.
⏳ 時間線
2023-04
Tongyi Qianwen (Qwen) Beta Launch
2024-06
Qwen 2 Series Released with 72B Flagship
2024-09
Qwen 2.5 Launch with Enhanced Reasoning
2025-04
Qwen 3 Family Debut (Apache 2.0 License)
2026-02
Qwen 3.5 and 397B MoE Open-Weight Release
2026-03
Qwen 3.5-Max-Preview Deployed on LMSYS Arena
📎 來源 (7)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： SCMP Technology ↗
每週電子報
每週一封，可隨時退訂。