Alibaba releases Qwen 3.8-Max AI model to challenge Fable 5 and GPT Sol
Alibaba Group has released Qwen 3.8-Max, a multimodal artificial intelligence model built on 2.4trln parameters with 95bn active parameters. The system is designed to compete directly with Anthropic’s Claude Fable 5 and OpenAI’s GPT Sol.
According to company disclosures, the model matches or outperforms US competitors on several benchmarks while utilising an open-source framework.
Priced at $2 per one mn input tokens and $6 per one mn output tokens, the model provides a commercial alternative to Western offerings. Subscription options are also available through Alibaba’s Token Plan starting at $6 per month. The company confirmed that the model weights will be released publicly next week to allow customisation. Following the announcement, Alibaba shares rose by up to 7.3% on the Hong Kong stock exchange.
Benchmark performance and task execution
Alibaba states that Qwen 3.8-Max improves on its predecessor across three core functional areas:
- Programming: The model executes full-stack software development tasks and complex coding assignments. Independent testers noted it handles code generation comparably to leading Western architectures.
- Office tasks and research: It processes long-duration administrative assignments, document analysis and structured research workflows.
- Multimodal processing: The architecture accepts text, images, video and large document inputs simultaneously, utilising a context window approaching one million tokens.
While Alibaba reports that its internal evaluations show performance comparable to Fable 5, independent third-party benchmarks have not yet verified these claims. The release follows a series of high-profile product announcements from Chinese technology firms competing in the domestic generative AI sector. Industry analysts note that Chinese developers are closing the performance gap with Silicon Valley despite ongoing restrictions on advanced semiconductor imports.
