Alibaba (Qwen)

Alibaba announced the establishment of the DAMO Academy (Discovery, Adventure, Momentum and Outlook), committing $15 billion over three years to fundamental technology research including AI, quantum computing, and fintech.

Alibaba Cloud announced Tongyi Qianwen, its large language model, positioning it as a competitor to ChatGPT and other LLMs. The model was designed for enterprise integration across Alibaba's cloud services.

Alibaba open-sourced Qwen-7B and Qwen-14B models, making them available for commercial and research use. This established Qwen as a major player in the open-weight LLM ecosystem.

Alibaba Cloud open-sourced Qwen-72B, a 72-billion parameter model that achieved state-of-the-art performance on Chinese language benchmarks. The release marked Alibaba's commitment to the open-source model ecosystem and established Qwen as the leading open-weight Chinese LLM family competing with Llama and GLM.

Alibaba released the Qwen1.5 model series with sizes from 0.5B to 110B parameters, alongside specialized variants Qwen-Audio, Qwen-Coder, and Qwen-VL for multimodal tasks. The expansion demonstrated Alibaba's strategy of offering specialized open-source models covering vision, audio, and code domains under a unified architecture.

Alibaba Cloud launched the Qwen-Max API, with its flagship model achieving top-tier rankings on Chatbot Arena and Chinese language understanding benchmarks. The release positioned Alibaba as a leading provider of API-access language models in China, competing directly with Baidu's ERNIE and Zhipu's GLM.

Alibaba released Qwen2, a significant upgrade with improved multilingual capabilities, coding performance, and mathematical reasoning. The models were open-sourced under Apache 2.0 license.

Alibaba released Qwen2.5, a family of models ranging from 0.5B to 72B parameters with improved coding, math, and reasoning capabilities, along with multimodal vision-language models.

Alibaba released Qwen3, the next generation of models with significant improvements in reasoning, coding, and multilingual performance, further solidifying Qwen's position as a top open-source LLM family.

Alibaba released Qwen3-Max, the largest and most capable model in the Qwen3 family. The preview version ranked third on the Text Arena leaderboard, surpassing GPT-5-Chat, and the official release enhanced coding and agent capabilities with state-of-the-art results across comprehensive benchmarks.

Alibaba's Qwen3.5-Max-Preview debuted on LM Arena, scoring ~1,464 Elo and ranking #6 globally at the time. This marked a significant capability jump for the Qwen family, though the model was released as closed weights with no public API, signaling a shift in Alibaba's open-weight strategy for its largest models.

Alibaba unveiled Qwen3.7-Max at the 2026 Alibaba Cloud Summit, the flagship proprietary model in the Qwen3.7 family. With a 1M token context window, extended thinking mode, and claims of autonomous operation up to 35 hours on long-horizon agentic tasks, it scored 57 on the Artificial Analysis Intelligence Index (first place on the public leaderboard) and ~1,475 Elo on LM Arena.

Alibaba unveils Qwen3.8-Max-Preview, a 2.4 trillion-parameter multimodal model, at the World AI Conference in Shanghai. The model is available through Token Plan, Qoder, and QoderWork. Alibaba describes it as second only to Claude Fable 5 among the models it benchmarked. Open weights promised for future release.

Alibaba officially launches Qwen 3.8-Max, a 2.4 trillion parameter open-weight model, fulfilling the preview announced at the World AI Conference in Shanghai (July 19). The model ranks competitively against frontier models from OpenAI and Anthropic, continuing Chinas aggressive push in the open-weight AI race with per-token pricing significantly below US competitors.

Alibaba's Qwen team released Qwen3.8-27B as an open-weight 27 billion parameter dense model under Apache 2.0 license. Achieves an Artificial Analysis Intelligence Index score of 52, tying GPT-5.6 Luna at max reasoning, and beats Opus 4.6 Max on SWE-Bench Pro. Runs at 68 tokens/second on a single NVIDIA RTX 4090 GPU (24GB), bringing frontier-level intelligence to consumer hardware for the first time. Supports 262K token context window with native vision-language capabilities.

Alibaba Group raised approximately $10.2 billion (HK$80 billion) through a secondary share placement in Hong Kong, the city's largest-ever secondary share sale, with all proceeds earmarked for AI infrastructure investment. The move underscores Alibaba's aggressive push to compete in the global AI race, though the stock dropped nearly 10% on the announcement day as investors questioned whether tech revenues can keep pace with AI spending.

Alibaba's Qwen team released Qwen3.8-Flash-Next, an open-weight multimodal mixture-of-experts model previewing the Qwen4 architecture, with 125 billion total parameters and 6 billion active per token. The model supports a 262,144-token context window, extendable to 1 million tokens with YaRN length extrapolation. Weights, the technical report, and inference code were published on ModelScope and GitHub under the Qwen Community license.

Qwen's first omni model designed for agentic use processed audio and video together across a 1-million-token context, edited vlogs, translated short videos, and summarized movies via tool use. API pricing at 0.15 USD per million input tokens undercut Google's Gemini 3.8 Flash (0.75 USD) by 80 percent while roughly matching its multimodal benchmarks, and open weights plus the open-source Qwen-MM-Plugins and Qwen-Live Harness shipped on GitHub the same day.

Alibaba's Qwen team released Qwen-Image-2.1 under open weights, a 7B-parameter single-stream diffusion transformer combining text-to-image generation, multi-reference editing of up to 10 images, and native RGBA transparent-image output in one unified model. Built on a Qwen3-VL 8B text encoder with a 64-channel RGBA autoencoder, it topped open-source image-generation leaderboards and matched or beat closed frontier models on standard benchmarks, extending Qwen's open-weight strategy from language into visual generation.