Alibaba unveils Qwen3.8-Max, its largest AI model yet, at 2.4 trillion parameters

Alibaba unveiled Qwen3.8-Max on August 3, 2026, the largest model in its Qwen family to date, with the company saying its benchmark scores are competitive with Anthropic's frontier models. The model is live now through Alibaba Cloud's Model Studio API for developers worldwide, with open weights due to follow the week of August 10 — the first time Alibaba has open-sourced a model at its Max tier.

What's new

• 2.4 trillion total parameters, using a Sparse Mixture-of-Experts architecture that activates roughly 95 billion parameters per token, keeping inference costs down despite the model's size.
• 1 million token context window — the model "supports a context window of up to 1 million tokens, which means it can understand and work with thousands of pages of information."
• Alibaba says the model delivers comparable or sometimes better scores than Anthropic's Fable 5 on its internal evaluations, though it still ranks fifth on the Text Arena leaderboard and second (behind Fable 5) on the Vision Arena leaderboard.
• Capabilities span coding, long-horizon agentic work, research, and visual intelligence — Alibaba highlighted use cases like reviewing legal documents, financial research, and interpreting hundred-page documents or 100-hour livestreams.
• In one internal test, the model reportedly worked autonomously for 16 days straight, writing, testing, and refining code on a software engineering project with minimal human intervention.
• API pricing undercuts Alibaba's own domestic rival: Qwen3.8-Max is priced below Moonshot AI's Kimi K3, which currently holds the title of China's largest model at 2.8 trillion parameters.

Context

Qwen3.8-Max arrives just weeks after Moonshot AI's Kimi K3 release, continuing a rapid back-and-forth between Chinese labs for the country's largest and most capable open AI system. It's also the fourth flagship-class Qwen release in 2026 following Qwen3.5, 3.6, and 3.7 — but the first at the Max tier that Alibaba has committed to open-sourcing, a departure from earlier Max-class models that stayed API-only. Alibaba's Hong Kong-listed shares rose 7% and its New York-listed shares climbed 4.5% the day of the announcement.

Why it matters

The release keeps pressure on the price-performance frontier that Chinese labs have been pushing since DeepSeek's early breakthroughs: a model Alibaba claims is competitive with Anthropic's best, priced well below both Anthropic's rates and a same-country rival's, and headed for an open-weight release within days. If the open-weight version matches the announced API specs, it hands developers unrestricted access to a model in the same performance class as closed frontier systems — intensifying cost competition for the coding and long-horizon agentic workloads that Western labs have built their commercial strategies around.