A naukri.com initiative
VentureBeat
3w
433
MiniMax-M1 is a new open source model with 1 MILLION TOKEN context and new, hyper efficient reinforcement learning
- Chinese AI startup MiniMax has launched MiniMax-M1, a large language model with 1 million input tokens and open-source under Apache 2.0 license.
- MiniMax-M1 sets new standards in long-context reasoning, agentic tool use, and efficient compute performance.
- It distinguishes itself with a context window of 1 million input tokens, outperforming models like OpenAI's GPT-4o.
- MiniMax-M1 is trained using a highly efficient reinforcement learning technique, consuming only 25% of the FLOPs required by other models like DeepSeek R1.
- The model has two variants, MiniMax-M1-40k and MiniMax-M1-80k, with different output lengths.
- MiniMax-M1's training cost was $534,700, which is significantly lower compared to other models like DeepSeek R1 and OpenAI's GPT-4.
- It achieves high accuracy on mathematics benchmarks and excels in coding and long-context tasks, outperforming competitors on complex tasks.
- MiniMax-M1 addresses challenges for technical professionals by offering long-context capabilities, compute efficiency, and open access under Apache 2.0 license.
- The model supports structured function calling, packaged with a chatbot API, and provides deployment options using Transformers library.
- MiniMax-M1 presents a flexible option for organizations to experiment with advanced AI capabilities while managing costs and avoiding proprietary constraints.
Read Full Article
26 Likes
For uninterrupted reading, download the app