DeepSeek has unveiled its V4.1 Flash model and simultaneously slashed the prices of its Flash series API. During off-peak hours, cached inputs are priced at a mere 0.02 yuan per million tokens. Meanwhile, non-cached inputs and outputs are set at 1 yuan and 4 yuan, respectively. Notably, these prices double during peak hours. The new model outperforms V4 Pro across the board. DeepSeek intends to reroute some Pro requests to Flash after September 14, defying the industry convention that improved model capabilities should command higher prices. On the same day, Zhipu and MiniMax saw their stock prices close down by 10.34% and 8.98%, respectively. Their cumulative declines since September have reached 33.64% and 22.98%. Despite both companies surpassing their full-year revenue from the previous year in the first half of 2026 and reporting robust August ARR (Annual Recurring Revenue) figures, Jefferies has revised its revenue forecasts for 2026-2029 upwards and lowered its loss projections. Simultaneously, it has reduced Zhipu's target price and slashed the valuation multiple for its API business from 50 times 2026 ARR to 30 times.

Both Zhipu and MiniMax have pivoted from their original core businesses, now heavily relying on API calls and enterprise services. While this strategy fuels rapid revenue growth and economies of scale, revenue becomes highly susceptible to token pricing fluctuations. Consequently, ARR becomes volatile, shifting the market's focus to the quality of growth. Zhipu once managed to achieve simultaneous growth in both volume and pricing. However, with the密集 (intensive, here it's better to use "rapid" to fit the context) rollout of low-cost, high-throughput models like Flash, the premium for model capabilities has quickly diminished. This enables clients to switch models at a low cost. Therefore, Zhipu must ensure that the growth in API call volume outpaces the decline in prices and that inference costs decrease faster than the quoted prices.

Currently, valuation multiples for independent model companies are being continuously revised downwards. This is because market expectations across three key dimensions are no longer scarce, making it challenging to convert product advantages into long-term pricing power. Large firms can seamlessly integrate models into their ecosystems. In contrast, independent model companies must retain clients, expand call volumes, and improve profit margins amidst falling prices. Capital markets will no longer reward growth alone.