MiniMax Targets Model Cost-Efficiency, M3 Processes Trillion Tokens Daily, H3 Multimodal Model Imminent

On July 23, MiniMax held an analyst meeting at its Shanghai headquarters. The company's management said that the current primary goal is to pursue the best cost-efficiency for each model. The M3 model's daily processing volume has reached the trillion-token level, and the open platform/API revenue has achieved a healthy gross margin. The company has built its own computing capacity, expanding resources through domestic and international acquisitions and long-term leases, which were secured when the long-term rental prices for GPUs and AI servers were significantly lower than current levels.

The H3 multimodal model is in the final stages of preparation for release. It can understand text, images, video, audio, music, and other inputs, and output video, images, and audio content based on user prompts.

MiniMax's current market value is approximately $8 billion to $8.5 billion. The company needs to prove that model usage, commercial revenue, and inference costs can all improve simultaneously. Its commercialization ability, cash burn, and self-funding capacity require continued observation.