Xiaomi's MiMo team released the MiMo-V2.6 series on 22 September 2026, publishing three open-weight omnimodal models that immediately claimed the top of the Artificial Analysis Intelligence Index for open weights with a 46.32 score.
Pro, Flash and Pro-UltraSpeed
The Pro flagship is natively omnimodal, accepting text, image, video and audio inputs. Its stack combines a 681 million-parameter vision encoder, a 308 million-parameter AudioTokenizer and a 127 million-parameter audio patch encoder, all served with a one-million-token context window. Flash is a compact efficient variant, while Pro-UltraSpeed strips the model back for latency-sensitive workloads at roughly 20x the Pro's throughput.
Pricing and benchmarks
Xiaomi published API pricing at $0.14 / $0.28 per million input and output tokens for Flash, $0.435 / $0.87 for Pro, and $4.35 / $8.70 for Pro-UltraSpeed. On agentic tasks the Pro scored 53.1 on AutomationBench (versus Claude Opus 5's 50.3), matched Opus 5 at 31.6 on Agents' Last Exam and edged it on Terminal Bench 2.1 at 89.9 versus 89.1. The models are available through OpenRouter and MiMo's own AI Studio, Desktop and Code platforms.
Context
The launch adds to a busy week for foundation models. See Anthropic's Claude Opus 5.5 launch, Alibaba's Qwen3-8-Omni-Flash and Google's Gemini 3.8 Flash TTS.
Reporting based on coverage from SiliconANGLE, eWeek and Times of AI.