Qwen Code Hints at Qwen4’s Million-Token Window

The breadcrumb is compelling, but developers should treat it as an internal readiness signal—not a confirmed model launch or guaranteed context length.

• –The change raises Qwen4’s auto-compaction threshold to roughly 850,000 tokens while leaving the output ceiling at 32,768.
• –`qwen4-omni-120b-think` suggests a multimodal reasoning model, though its presence in test fixtures does not prove public availability.
• –The PR author says testing used a mock endpoint and that the 1M figure came from reviewer guidance rather than measurement.
• –If accurate, Qwen4 would enter the long-context race with a major advantage for repository-scale coding and multimodal workflows.
• –Qwen Code is an open-source terminal coding agent, making the repository a meaningful early signal for developers watching Alibaba’s model roadmap. ([official repository](https://github.com/QwenLM/qwen-code))