MiniMax shares jumped 19.3% in Hong Kong.

Recently, MiniMax officially announced the open-sourcing of OctoCodingBench, a systematic evaluation benchmark for coding agents developed in-house. It is the industry’s first comprehensive benchmark specifically designed for Coding Agents. Evaluation results show that on the key metric of process compliance, some open-source models have delivered outstanding performance, rapidly narrowing the gap with — and in certain scenarios even surpassing — proprietary models. This development suggests that as AI evolves toward the agent era, data quality and evaluation paradigms, rather than standalone model performance, are becoming the new focal points of industry competition.

According to a research note from China Securities Construction Investment (CSCI), amid the global surge of generative AI, MiniMax has distinguished itself by adhering to a “counter-consensus” strategy that prioritizes breakthroughs in model intelligence. As one of the first companies in Shanghai to receive regulatory approval for large models, MiniMax demonstrates strong growth potential through deep technical expertise and forward-looking commercialization capabilities. The firm forecasts that from 2025 to 2027, MiniMax’s revenue will grow at a CAGR exceeding 90%, with Non-GAAP gross margin expected to rise to 55% and net loss margins continuing to narrow. With ongoing inference cost optimization and the rollout of next-generation multimodal models, MiniMax is well positioned to unlock broader opportunities in AI-native applications.