OpenAI’s 21B-parameter open-weight Mixture-of-Experts (MoE) model, released under the Apache 2.0 license. Features 3.6B active parameters per forward pass, optimized for low-latency inference and deployability on consumer or single-GPU hardware. Trained in OpenAI’s Harmony response format, it supports reasoning level configuration, fine-tuning, and agentic capabilities such as function calling and structured outputs.
MiMo-V2.6-Flash is an open-source foundation model from Xiaomi. It uses a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token, plus a hybrid attention mechanism for better compute efficiency. It has a 1M-token context window and native multimodal support, and is tuned for agentic workflows in coding, visual, general, and research scenarios, including long-horizon tasks with different agent harnesses.