Chinese LLM APIs in 2026: Flagship Models Cost ¥4–¥12 Per Million Tokens

As of August 21, 2026, Chinese large language model APIs offer some of the most competitive pricing globally, with flagship models ranging from ¥4.00 to ¥12.00 per million input tokens, according to pricing data verified by llmabacus. Leading models in this range include Baidu ERNIE 5.1 at the low end and Alibaba's Qwen3.7 Max at the high end, while budget options like Qwen3.5 Flash go as low as ¥0.20 per million input tokens. DeepSeek's models stand out for aggressive caching discounts, with DeepSeek V4 Flash offering cached input at just ¥0.10 per million tokens. Analysts attribute the sustained price decline to cheaper hardware, intensifying domestic competition, and the rise of aggregator gateways that unify access across vendors. Compared to Western counterparts, Chinese value-tier models are estimated to be 80–98% cheaper than GPT-5.5-class equivalents, though buyers are advised to weigh cache hit rates and tool-calling compatibility alongside list prices.