Alibaba's Qwen 3.8 Max trails Claude Fable, surpasses GPT-5.6 in coding
New open-source model matches proprietary leaders at 90% lower inference cost.
Deep Dive
This video covers Qwen 3.8 Max—from performance and benchmark scores to coding abilities, reasoning improvements, speed, pricing, and how it stacks up against GPT-5.6, Claude Fable, Gemini, DeepSeek, and other leading AI models.
Key Points
- Qwen 3.8 Max achieves 94% on HumanEval, beating GPT-5.6 (91%) and DeepSeek-V3 (89%).
- Priced at $0.50 per million tokens, it costs 10x less than GPT-5.6 for comparable performance.
- Supports 128K context and 3x faster inference via MoE architecture with 16 experts.
Why It Matters
Frontier open-source AI at 10% cost threatens proprietary margins, democratizing advanced reasoning for enterprises.