Alibaba's Qwen3.5-397B-A17B: Massive Open-Source Beast Hits GPQA 0.9!
A 397-billion parameter open-source AI model achieves near-expert performance on a PhD-level benchmark.
Alibaba Cloud's Qwen Team has made a major contribution to the open-source AI landscape with the release of Qwen3.5-397B-A17B on February 16, 2026. This massive 397-billion parameter model represents a significant leap in accessible high-performance AI, achieving a score of 0.9 on the challenging GPQA benchmark. The launch is part of an explosive wave of new model releases this month, intensifying competition and accelerating the pace of innovation in the foundation model space. By open-sourcing such a capable model, Alibaba is directly challenging the dominance of proprietary models from companies like OpenAI and Anthropic, providing a powerful tool for the global research and developer community.
The model's standout achievement is its performance on the GPQA (Graduate-Level Google-Proof Q&A) benchmark, where a score of 0.9 indicates near-expert-level reasoning and knowledge recall in specialized scientific domains typically requiring PhD-level understanding. This positions Qwen3.5-397B-A17B as one of the most capable open-source models available for complex reasoning tasks. The release strategy underscores a growing trend where major tech firms are leveraging open-source models to build ecosystem influence and drive adoption of their cloud platforms. For developers, this means unprecedented access to a model that can power advanced applications in research, data analysis, and enterprise AI without the constraints and costs of closed APIs, fundamentally altering the competitive dynamics of the AI industry.
- Alibaba's Qwen Team released the 397-billion parameter Qwen3.5-397B-A17B model on February 16, 2026.
- The model scores 0.9 on the GPQA benchmark, demonstrating near-expert performance on PhD-level questions.
- This open-source release is part of a major wave of new models, increasing access to high-performance AI.
Why It Matters
Democratizes access to near-expert-level AI reasoning, challenging proprietary models and empowering global developers.