The rollout caps a rapid, open-weight strategy for Qwen, which began as Tongyi Qianwen in 2023 and has accelerated toward 2.4 trillion open weights by 2026, underpinned by hybrid thinking and multilingual expansion.

In 2023, the project kicked off with Tongyi Qianwen’s April reveal, a public unveiling at Alibaba Cloud Summit, and open-sourcing Qwen-7B and Qwen-7B-Chat in August, followed by Qwen-VL and broad public access by September, concluding with 72B and 1.8B models in December to cover a spectrum from portable devices to frontier-scale models.

The open-weight family expanded in 2024, introducing Qwen1.5 in February to diversify models and contexts, Qwen2 in June with five sizes and multilingual expansion including a MoE flagship, and Qwen2.5 in September, which released more than 100 open-source models and increased training tokens to 18 trillion, plus Qwen2.5-Coder and QwQ-32B/QVQ-72B-Preview adding multimodal and reasoning capabilities.

Early 2025 showcased multimodal progress and scaled MoE options, with Qwen2.5-VL for PC/phone control and Qwen2.5-Max as a large MoE, followed by milestones like QwQ-32B and Qwen 3-Omni/VL in March.

In April 2025, Qwen3 debuted with a 36-trillion-token, 119-language family, and by September its API-only flagship and large-scale hybrids (Qwen3-Max and related variants) marked significant progress toward institutional adoption, including Singapore’s Sea-Lion program positioning Qwen as infrastructure.

The licensing arc shifted toward a two-tier model: open for small/mid-size adoption, API/proprietary for frontier use, unlocked by a 2.4T open-weight threshold tied to revenue, with over 400 open models and 1 billion downloads by March 2026; the Qwen app reached 203 million monthly active users by February 2026.