Alibaba takes aim at OpenAI and Anthropic with Qwen3.8-Max launch

Alibaba has unveiled Qwen3.8-Max, the company's largest AI model to date, aimed at the enterprise market. This 2.4-trillion-parameter mixture-of-experts (MoE) model is designed to excel in software engineering, multimodal reasoning, and other knowledge-intensive tasks, while offering improved inference efficiency. Qwen3.8-Max activates only about 95 billion parameters during inference, making it more cost-effective than its larger counterparts.

Alibaba claims Qwen3.8-Max outperforms leading AI models such as Anthropic's Claude Opus 4.8, Claude Fable 5, and OpenAI's GPT-5.6 Sol on coding benchmarks, including SWE-bench Pro and a proprietary NL2Repo-Bench. The model has undergone thorough testing, with one project reportedly completing a software engineering task in 16 days without any human intervention. This includes projects spanning legal compliance, financial analysis, engineering design, quantitative research, and multimodal content creation.

While Alibaba highlights the reduced inference costs due to its MoE architecture, industry analysts emphasize that enterprise buyers may prioritize evaluation throughput and cost considerations over raw model size. Open-weight models like Qwen3.8-Max provide greater flexibility for customization and deployment, but enterprises should carefully evaluate the lack of indemnification protections and the need for security, governance, and code-scanning controls to mitigate potential risks.

CIOs are advised to prioritize transparency and total cost of ownership when evaluating Qwen3.8-Max, as the model's performance and suitability will depend on specific organizational requirements and infrastructure.