MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention

We introduce MiniMax-M1, the world's first open-weight, large-scale hybrid-attention reasoning model. MiniMax-M1 is powered by a hybrid Mixture-of-Experts (MoE) architecture combined with a lightning attention mechanism. The model is developed based on our previous MiniMax-Text-01 model, which contains a total of 456 billion parameters with 45.9 billion parameters activated per token. The M1 model natively supports a context length of 1 million tokens, 8x the context size of DeepSeek R1. Furthermore, the lightning attention mechanism in MiniMax-M1 enables efficient scaling of test-time compute. These properties make M1 particularly suitable for complex tasks that require processing long inputs and thinking extensively. MiniMax-M1 is trained using large-scale reinforcement learning (RL) on diverse problems including sandbox-based, real-world software engineering environments. In addition to M1's inherent efficiency advantage for RL training, we propose CISPO, a novel RL algorithm to further enhance RL efficiency. CISPO clips importance sampling weights rather than token updates, outperforming other competitive RL variants. Combining hybrid-attention and CISPO enables MiniMax-M1's full RL training on 512 H800 GPUs to complete in only three weeks, with a rental cost of just $534,700. We release two versions of MiniMax-M1 models with 40K and 80K thinking budgets respectively, where the 40K model represents an intermediate phase of the 80K training. Experiments on standard benchmarks show that our models are comparable or superior to strong open-weight models such as the original DeepSeek-R1 and Qwen3-235B, with particular strengths in complex software engineering, tool utilization, and long-context tasks. We publicly release MiniMax-M1 at https://github.com/MiniMax-AI/MiniMax-M1.

ChatGPT as the Transportation Equity Information Source for Scientific Writing

Transportation equity is an interdisciplinary agenda that requires both transportation and social inputs. Traditionally, transportation equity information are sources from public libraries, conferences, televisions, social media, among other. Artificial intelligence (AI) tools including advanced language models such as ChatGPT are becoming favorite information sources. However, their credibility has not been well explored. This study explored the content and usefulness of ChatGPT-generated information related to transportation equity. It utilized 152 papers retrieved through the Web of Science (WoS) repository. The prompt was crafted for ChatGPT to provide an abstract given the title of the paper. The ChatGPT-based abstracts were then compared to human-written abstracts using statistical tools and unsupervised text mining. The results indicate that a weak similarity between ChatGPT and human-written abstracts. On average, the human-written abstracts and ChatGPT generated abstracts were about 58% similar, with a maximum and minimum of 97% and 1.4%, respectively. The keywords from the abstracts of papers with over the mean similarity score were more likely to be similar whereas those from below the average score were less likely to be similar. Themes with high similarity scores include access, public transit, and policy, among others. Further, clear differences in the key pattern of clusters for high and low similarity score abstracts was observed. Contrarily, the findings from collocated keywords were inconclusive. The study findings suggest that ChatGPT has the potential to be a source of transportation equity information. However, currently, a great amount of attention is needed before a user can utilize materials from ChatGPT