Quick answer
September 25, 2026: Alibaba cut prices across its Qwen audio models, with speech recognition (ASR) down 95%, text-to-speech down about 70%, and the realtime audio API down about 85%, alongside new TTS-Next and ASR-Next models. The same day DeepSeek, reported to have reached a $1 billion annual revenue run-rate, raised its API prices by between 2.3 and 4.5 times depending on the model. The two moves point in opposite directions for a reason: audio is a land grab, and DeepSeek's text models no longer need to be loss leaders.
For two years the story of Chinese AI pricing was one direction: down. September 25 broke the pattern in an instructive way.
Qwen: audio gets nearly free
Alibaba's cuts make its speech recognition among the cheapest available anywhere, and its text-to-speech and realtime audio competitive with the lowest Western prices. The new TTS-Next and ASR-Next models were announced with the cuts, which suggests the price drop is partly a launch strategy: get developers onto the new models before rivals respond. For voice agents, transcription products, and dubbing, the transcription and synthesis line items just shrank.
DeepSeek: the discount ends
DeepSeek's January 2025 launch reset expectations for what a frontier-class model should cost, and its prices stayed far below Western rivals since. A rise of 2.3 to 4.5 times, reported alongside a $1 billion run-rate, says the company now has the demand to price closer to the market. Even after the rise, its models remain cheaper than most Western equivalents, but the era of near-free frontier text inference from DeepSeek is over.
What developers should do
- Voice products: re-quote your audio costs; Qwen's cuts may change the provider decision
- DeepSeek users: check the new rates against Qwen, Grok 4.7, and the September Western launches; the gap has narrowed sharply
- Everyone: hosted prices move both ways now; build with a gateway so a provider switch is a config change
- Note the usual caveats on Chinese-hosted services for data residency and content policy
Related reading
Bottom line
Audio is the new price war; text is maturing. Qwen's cuts are worth a re-quote for anyone building voice. DeepSeek's rise is a sign the cheapest frontier model is now a business, not a statement.
