Qwen

Qwen is the family of large language and multimodal models made by Alibaba, which says Qwen powers the AI behind its enterprise and consumer services.¹ The Qwen team first published weights on August 3, 2023, releasing Qwen-7B and Qwen-7B-Chat on ModelScope and Hugging Face,² and its technical report, posted September 28, 2023, called Qwen "the first installment" of a series spanning base, chat, coding and math models.³ In February 2026 Alibaba open-sourced Qwen3.5-397B-A17B, a natively multimodal model released under Apache 2.0, and in May 2026 it released Qwen3.7-Max for agentic workloads.¹⁴ Alibaba Cloud's current API lineup is led by qwen3.8-max and the low-cost qwen3.8-flash, each priced for requests of up to 1M input tokens.⁵

Qwen's reach is measured in Chinese Open-Weight Models: Downloads, Derivatives and Token Share. The ATOM Report counts 942.1 million cumulative Hugging Face downloads for Qwen by March 2026, passing Meta's Llama in September 2025, and finds Qwen's share of new fine-tunes rising from 1% in January 2024 to 69% in February 2026.⁶ Hugging Face counts more than 113,000 derivative models of the Qwen family, and over 200,000 models that tag Qwen.⁷ Small models weigh heavily in those counts, the ATOM authors note; Qwen3-0.6B, Apache 2.0, logged 30,726,361 downloads in the 30 days to October 13, 2026.⁶⁸ On price, qwen3.8-flash ranks first in Cheapest LLM APIs per 1M Output Tokens at $0.47 per 1M output tokens.⁵ In speech, Qwen3-ASR-1.7B averages a 5.40% word error rate on the Open ASR Leaderboard, as covered in What Word Error Rate Measures, and Why Speech-to-Text WERs Disagree.⁹¹⁰