Alibaba's Qwen Code repository inadvertently revealed upcoming models including qwen4-max, qwen4-coder-plus, and qwen4-omni-120b through pull requests. The leaked code references model identifiers such as qwen4, qwen4-plus, qwen4.5-max, and qwen4-max-2026-10-01, all configured with a 262,144 context window. Notably, qwen4-omni-120b-think is already implemented as the session model in omni fixtures, confirming these models are beyond hypothetical stages.
Read original
reddit/r/LocalLLaMA