RIZOHahaYes Price

On August 26, MiniMax disclosed during its mid-2026 performance meeting that its annual recurring revenue (ARR) for August has surpassed $800 million, with a 81.8% quarter-over-quarter increase in revenue from the second quarter compared to the first quarter. The Token consumption in July reached 20 times that of January, and the contribution of its B-end business to the overall ARR has exceeded 80%. MiniMax stated that in terms of computing power, it is one of the first two independent large model companies in the country to establish a scaled, long-term stable infrastructure system. The infrastructure is now capable of independently supporting the largest training tasks, with the Effective Training Time Ratio (ETTR) reaching 97%.

On August 26, MiniMax disclosed during its mid-2026 performance meeting that its annual recurring revenue (ARR) for August has surpassed $800 million, with a 81.8% quarter-over-quarter revenue growth from the first to the second quarter. The token consumption in July reached 20 times that of January, and the contribution of its B-end business to the overall ARR has exceeded 80%. MiniMax stated that in terms of computing power, it is one of the first two independent large model companies in the country to establish a large-scale, long-term stable infrastructure system. This infrastructure is now capable of independently supporting the largest training tasks, with an Effective Training Time Ratio (ETTR) reaching 97%.

On August 26, Alibaba Qwen launched the Qwen 3.8-Flash model, which is a multimodal MoE model and an early preview of the Qwen 4 architecture. The production version of Qwen 3.8-Flash will soon be available through the Qwen Cloud API, priced at just $0.16 per million input tokens and $0.47 per million output tokens. The model features 125 billion parameters plus 51 billion N-gram embedding parameters, but activates only 6 billion parameters per token, achieving high cost-effectiveness.