anagnorisis.cloudSign in

← Hourlies

Hourly ·

Alibaba Open-Sources Qwen3.8-Max, Its Most Capable Model, as Downloads Pass 3 Billion

Alibaba is open-sourcing Qwen3.8-Max, its most capable model yet — a 2.4-trillion-parameter system whose weights land on Hugging Face and ModelScope next week — as the Qwen family passes 3 billion downloads and overtakes Meta and Google in open-weight AI.

Alibaba Open-Sources Qwen3.8-Max, Its Most Capable Model, as Downloads Pass 3 Billion
Image credit: Software: Alibaba CloudScreenshot: VulcanSphere, Public domain (license)

Alibaba is giving away its most powerful AI model for free. The company released Qwen3.8-Max — the first "Max"-class Qwen model to ship as open weights — with the full parameters landing on Hugging Face and ModelScope next week at no cost.

The scale is substantial: 2.4 trillion total parameters with 95 billion active at any given moment, a mixture-of-experts design meant to run efficiently on modest hardware. The model ships with setup instructions for Anthropic's Claude Code and OpenAI's Codex, rival coding tools, so developers can slot it into either workflow.

By Alibaba's own benchmark scorecard, Qwen3.8-Max doesn't top the text charts. Across 31 text tests, Anthropic's Fable 5 wins 15, OpenAI's GPT-5.6 Sol takes nine, and Qwen takes seven — and just one of 12 coding tests. But the rankings invert on multimodal work, from documents to video to spatial reasoning, where Qwen leads most of the table. The pitch is economics: at about 30% of what Claude Fable 5 charges, second place is cheap enough to matter.

The strategy is distribution over raw dominance. Alibaba says the Qwen family has passed 3 billion downloads, making it the most-used open-weight line in the world, ahead of anything from Meta or Google. Hugging Face's own State of Open Models report, published a day earlier, counts a smaller but still decisive 2.045 billion this year (2.061 billion including every repository) and 151,448 Qwen-based derivatives — roughly 2.6 times Meta's entire footprint on the Hub.

The backdrop is a widening split in the AI race. Chinese open-weight models went from under 2% of tokens on OpenRouter in late 2024 to roughly 61% by mid-2026, and Qwen has overtaken Meta's Llama as the world's most self-hosted model. With Washington restricting Fable 5 and Mythos 5 under export controls in June, and Beijing weighing its own limits on Chinese models going overseas, Alibaba is winning on reach even when it doesn't top the leaderboard.

Sources: Decrypt | The Next Web | Global Times

More Hourlies Stories

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.

More from Anagnorisis