Unsloth released a 2-bit GGUF version of MiniMax-M3.1-Flash, reducing the model size to 143.9GB, about 45% smaller than the original, while retaining 91.5% top-1 token agreement in Unsloth's evaluation. The version supports local deployment, llama.cpp, and an OpenAI-compatible API to lower self-hosting barriers.
Inferact announced a strategic partnership with MiniMax to advance production inference for open-source models, with engineering improvements contributed back to vLLM. The Inferact Platform has served MiniMax M3.1 Flash since its launch, and MiniMax platform requests are routed to Inferact.
MiniMax announced the open-sourcing of its MiniMax M3.1 Flash model weights, with 428B total parameters and 23B activated parameters. The model launched on October 19 and processed over 100 trillion tokens within 72 hours. MiniMax is partnering with chip vendors including Huawei Ascend, NVIDIA, and AMD, along with multiple cloud platforms, for day-one integration, with communities such as vLLM and Hugging Face supporting ecosystem adoption.
Yuangu Tiantan, a new media account under China Media Group, revealed on October 22 that DeepSeek and MiniMax have completed registration procedures and finished open-sourcing and overseas licensing since the Ministry of Commerce and the Ministry of Science and Technology solicited public comments on revising the Catalogue of Technologies Prohibited or Restricted from Export on October 16.
On October 20, 2026, Jefferies maintained a Buy rating and HK$533 target price on Minimax, citing the release of the M3.1 Flash model with significant capability gains in agentic tasks, programming, computer use, and cybersecurity, plus roughly 3x faster inference and 70% lower costs.
On October 19, MiniMax released the M3.1 Flash model, cutting costs on agentic tasks by 70% compared to the previous generation. API pricing is RMB 1 per million input tokens, RMB 4 per million output tokens, and RMB 0.1 per million cache reads. Model weights will be open on October 23.
MiniMax released its new Flash model MiniMax M3.1 Flash, achieving generational improvements in agentic tasks, coding, computer use, and cybersecurity, with nearly 3x faster online inference and 70% lower usage costs. Input is priced at 1 yuan per million tokens, output at 4 yuan, and cache read at 0.1 yuan. It is now available on MiniMax API, MPlan, and MiniMax Code, with weights to be released later.
MiniMax officially claimed ownership of the anonymous model Space Bunny via a short video on October 19, 2026. The model topped daily token call volume on OpenRouter and OpenCode within three days of its late-September launch, ranking alongside DeepSeek V4.1 Flash and GLM 5.3 Flash.
Reactor introduced FastH3-Omni, an open-source realtime audio-video generation model based on MiniMax's H3 model and post-trained by FastVideo and NuvA Lab. It supports first/last frame conditioning and audio input, with an exclusive preview on the Reactor API until October 20, 2026.
MiniMax announced that its latest text model M3.1-Flash-Preview has been launched on the MiniMax Code platform, providing stable productivity for daily development, from bug fixing to full feature development. The company also issued a quota reset card to reset all users' Token Plan quotas, and introduced benefits such as double check-in points and free tokens.