Understand this story
On October 26, 2026, Unsloth released a 2-bit GGUF quantized version of MiniMax-M3.1-Flash, reducing its size by 45% while retaining 91.5% top-1 accuracy, runnable on 150GB of memory, and provided a GGUF download link and usage guide. The next day, MiniMax announced that M3.1 Flash had been integrated into the Unsloth community; its 428B MoE model, after Unsloth dynamic GGUF quantization, was reduced to a 2-bit footprint of 143.9GB, deployable via Unsloth Desktop and llama.cpp, with an OpenAI-compatible API provided. Separately, according to TMTPost, MiniMax released M3.1 Flash on October 19, and Artificial Analysis gave it a composite intelligence index score of 42, 13 points higher than M3 in June, but it was subsequently questioned for memorizing answers in evaluations; the model processed over 100 trillion tokens within 72 hours of launch, and its model weights were opened on October 23.
AI-generated from public reports
Latest development · Report
MiniMax M3.1 Flash Faces Memorization Doubts After Low-Price Leaderboard EntryDevelopments
2 confirmed developments
- Source:TMTPost · AGI
Unsloth releases locally runnable MiniMax-M3.1-Flash 2-bit GGUF
Source:Unsloth
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 27
TMTPost · AGI
MiniMax M3.1 Flash Faces Memorization Doubts After Low-Price Leaderboard Entry
MiniMax released M3.1 Flash on October 19. Artificial Analysis gave it a composite intelligence index of 42, 13 points higher than M3 from June, but it was subsequently accused of memorizing questions in evaluations. Evaluation blogger Mao Laoban said the model scored near perfect on retained old questions and near zero on newly added questions; after blogger Binkg changed the conditions of a production management question, DeepSeek V4.1 Flash-1017 and Gemini 4 Argon both changed their answers to RCCP, while M3.1 Flash still answered RRP. A TMTPost retest found that it answered incorrectly when asked in English, but answered correctly when a separate conversation was started in Chinese. The model processed over 100 trillion tokens within 72 hours of launch; official API input and output prices are $0.15 and $0.60 per million tokens, respectively, and the model weights were opened on October 23.
TMTPost · AGIRelated discussion
TMTPost · AGI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
MiniMaxFirst-party
MiniMax M3.1 Flash Now Runs Locally via Unsloth Dynamic GGUF
MiniMax announced that M3.1 Flash has joined the Unsloth community; after Unsloth dynamic GGUF quantization, its 428B MoE model's size drops to a 143.9GB 2-bit footprint. Unsloth's evaluation shows that this quantized version maintains a 91.5% top-1 token agreement rate across seven domains, can be deployed via Unsloth Desktop and llama.cpp, and offers an OpenAI-compatible API. MiniMax also noted that Unsloth has 78K+ stars on GitHub and over 500M model downloads on Hugging Face.
Oct 26
UnslothFirst-partySelected
Unsloth makes MiniMax-M3.1-Flash runnable locally
Unsloth has released a 2-bit GGUF quantized version of MiniMax-M3.1-Flash, which retains 91.5% top-1 accuracy while reducing size by 45%, and can run on 150GB of memory. Unsloth claims this model is the strongest among open-source models of the same size, and provides a GGUF download link and usage guide. MiniMax officially previously introduced that M3.1 Flash reaches 62% on Automation-Bench, 74% on DeepSWE, and 91% on BabyVision, with API pricing at $0.15/$0.60 per 1M input/output tokens.
Heat trend for this story
- Current
- 12
- Period peak
- 22 Oct 27 20:00
- 24-hour change
- -9%
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.