MiMo-V2.6-Pro

Codersera's guide documents the MiMo-V2.6 family of four models released by Xiaomi on 22 September 2026 (China time, with Hugging Face repositories published late on 21 September UTC), five months after MiMo-V2.5. The flagship MiMo-V2.6-Pro is described as a 1.02T-total / 42B-active omnimodal sparse MoE with a 1M conte The guide reports that MiMo-V2.6-Pro scored 46 on Artificial Analysis's Intelligence Index versus V2.5-Pro's 26, placing it ahead of GLM-5.3, Kimi K3, and DeepSeek V4.1 Flash among open-weight flagships, with API pricing of $0.435 input and $0.87 output per million tokens. It also names two additional family members: M

Forkast reports Xiaomi released the MiMo-V2.6 series on 22 September 2026 with MiMo-V2.6-Pro scoring 46 on Artificial Analysis Intelligence Index v4.3, tying Grok 4.7 and setting a new ceiling for open-weight models. The article documents the model's frozen-router Mixture-of-Experts architecture at 1.02T total / 42B ac Vendor-reported benchmarks cited include 71.9% on DeepSWE v1.1 (trailing Claude Opus 5's 74.0% but ahead of Claude Fable 5), 89.9% on Terminal Bench 2.1 surpassing Claude Opus 5 and GPT-5.6 Sol, 94.0% on CyberGym leading all reported models, and 53.1% on AutomationBench v1.0.6 outpacing US counterparts. Both Pro and Fl

Latent Space's AINews roundup reports Xiaomi's release of MiMo-V2.6-Pro as the top open-weights model on Artificial Analysis's Intelligence Index at a score of 46, with $0.13 per Intelligence Index task cost placing it on the Intelligence vs. Cost Pareto frontier. The post quotes Xiaomi describing MiMo-V2.6-Pro as the The piece highlights Xiaomi's unusually transparent final RL training runs (published live by former DeepSeek engineer Luo Fuli) and details the three axes along which the team scaled RL compute: larger batches at up to 1M context (1,568 samples per update, 3.5–3.7B tokens per step), richer multi-task environments span

Kingy.ai's guide documents access options for MiMo-V2.6-Pro, confirming it is not free via API (priced at $0.435/$0.87 per million tokens) but noting the V2.6 series weights are free to download from Hugging Face under the MIT license, while MiMo-V2.6-Flash is offered free for about a week inside OpenCode's coding agen The guide also catalogs paid access routes: Xiaomi's OpenCode Go plan at $10/month includes both Pro and Flash among 30+ models with spending caps and 5-hour/weekly limits, while Xiaomi's Token Plan Lite starts at $6/month with 4.1 billion credits for using V2.6-Pro and Flash inside OpenCode, Claude Code, and other too

Prompt Blueprints reproduces Xiaomi's announced benchmark tables for MiMo-V2.6 Pro and Flash, noting both are described by Xiaomi as omnimodal models built via scaled reinforcement learning across coding, computer-use, 3D-reasoning, and creative capabilities, with the open release including weights, a technical report, The reproduced table records MiMo-V2.6-Pro scoring 71.9 on DeepSWE v1.1, 26.5 on ProgramBench, 63.2 on MiMo Code Bench, 53.1 on AutomationBench v1.0.0.6, 76.9 on Toolathlon-Verified, 1673 on GDPval-AA, 31.6 on Agents' Last Exam, 34.9 on Terminal Bench 4.0, and 89.9 on Terminal Bench 2.1, compared against MiMo-V2.6 Flas

Benchable's model page for MiMo-V2.6-Pro lists a 1M-token context window, support for text, image, video, and audio input producing text output, and pricing of $0.435 per 1M input tokens, $0.87 per 1M output tokens, and $0.0036 per 1M cached input tokens. Total parameters are listed as approximately 1T (noted as "Rumou Performance metrics on the page show MiMo-V2.6-Pro with 44th-percentile speed, 52nd-percentile pricing, and 99% reliability, with perfect scores in General Knowledge and Reasoning, 95% accuracy in Coding (88th percentile), 99% in Email Classification (85th percentile), 73% in Instruction Following (71st percentile), 94

Artificial Analysis published an intelligence, performance and price profile for Xiaomi's MiMo-V2.6-Pro, scoring the open-weights model 46 on its Intelligence Index, placing it well above the median of 18 among its 114-model comparison class. The profile lists pricing at $0.435 per 1M input tokens and $0.87 per 1M outp Technical specifications from the page describe MiMo-V2.6-Pro as a 1.0T total / 42B active parameter mixture-of-experts model with a 1M-token context window, MIT-licensed weights hosted on Hugging Face, and support for text, image, speech, and video input producing text output. The page flags that the profiled variant