Open-source seems to not only holding its own, but also making dents into the usage of popular proprietary AI models.
New data from Open Router, a popular platform for routing requests to various AI models, reveals a significant shift in the landscape of code generation models. Alibaba’s open-source Qwen3 Coder has made impressive gains, seemingly at the expense of proprietary models from Anthropic and Google.
Data from the week of July 14th shows Anthropic’s Claude Sonnet 4 leading the pack with a substantial 46.3% of the market share on Open Router. Google’s Gemini 2.5 Pro held a respectable 13.4%. Fast forward to the week of August 11th, and the picture looks remarkably different. While Claude Sonnet 4 still holds the top spot, its share has dropped to 32.3%. Google’s Gemini 2.5 Pro has also seen a decline, now sitting at 5.8%.
The most striking change, however, is the meteoric rise of Qwen: Qwen3 Coder, which has claimed a remarkable 20.7% of the market share in just a few weeks. This surge could be an indication that developers are not only open to using open-source models but are actively embracing them for their coding needs.
Qwen3-Coder-480B-A35B-Instruct is a Mixture-of-Experts (MoE) code generation model developed by the Qwen team. It is optimized for agentic coding tasks such as function calling, tool use, and long-context reasoning over repositories. The model features 480 billion total parameters, with 35 billion active per forward pass (8 out of 160 experts). The model is competitively priced, and at least going by OpenRouter data, seems to be gaining traction at the expense of popular coding models like Claude Sonnet 4 and Gemini Pro 2.5.
More impressively, Qwen-3-Code is open-source, unlike offerings from Google and Anthropic, and developers can modify it for their own particular needs. Developers thus likely believe they have more control over the model, as opposed to closed-source models, which are run entirely by large corporations, and as OpenAI recently showed, can be taken down at will. Even more interestingly, Qwen is made by a Chinese company, so a Chinese company is taking away market share from American labs. It remains to be seen if Qwen-3-coder can sustain the momentum, but with models like DeepSeek, Qwen, Kimi K2 and Z.ai in the fray, China is making a serious play in the AI race.