Stay Ahead of the Curve
Latest AI news, expert analysis, bold opinions, and key trends — delivered to your inbox.
Alibaba Drops Open-Weight Qwen3 AI: A Direct Challenge to OpenAI and DeepSeek
2 min read 8 new open-source models, hybrid reasoning, 119 languages—Alibaba’s bold new AI family could shake up the global leaderboard. May 04, 2025 16:19
Alibaba just launched Qwen3, its third-generation open-source AI model family—and it’s aiming straight at the big players. With eight models built for reasoning, coding, and multilingual tasks, Alibaba claims Qwen3 can match (or even beat) models like DeepSeek’s R1 and OpenAI’s o1.
🔓 Key Highlights
- Hybrid Reasoning:
All Qwen3 models feature two modes:
– “Thinking” mode for complex, logic-heavy queries (slower but smarter)
– “Non-thinking” mode for fast, everyday responses - Multilingual Support:
Each model supports 119 languages, expanding their usability across global contexts. - Massive Training Dataset:
Qwen3 was trained on 36 trillion raw data tokens—including textbooks, Q&A datasets, code, and synthetic AI-generated material. - Benchmark Performance:
Alibaba says the flagship model, Qwen3-235B-A22B, beats DeepSeek’s R1 and OpenAI’s o1 on key benchmarks. Other models also performed exceptionally well in math and coding.
🔐 Why It Matters
Despite U.S. efforts to limit China's access to advanced AI chips, Chinese firms like Alibaba and DeepSeek continue to push ahead. Qwen3 proves that China isn’t just catching up—it’s competing at the frontier.
Even more, it reflects a growing trend: companies building with open-weight models like Qwen3 instead of relying on closed systems from OpenAI or Anthropic.
If the benchmarks hold, Qwen3 could become a major force in the global AI race—especially in enterprise and multilingual applications.
AI Agents