DeepSeek warns of 'significant' API price hikes — plan ahead

DeepSeek dropped a notice on August 6 telling developers it plans to "broadly increase" API pricing in the near term, and that the increase is expected to be significant. No new rates. No effective date. Just: "Please plan your usage accordingly."

This is the second pricing move in under a month. Mid-July brought a peak/off-peak mechanism — roughly 2x during busy hours, discounts overnight. The new hike lifts the base, so even off-peak calls get more expensive. Current published rates sit around $0.14/$0.28 per million tokens for V4 Flash and $0.435/$0.87 for V4 Pro — the numbers that made DeepSeek the default "good enough and dirt cheap" backend for a lot of side projects and Chinese startups.

The subtext isn't subtle. Inference at that price point only works if you're subsidising it or you have spare compute, and DeepSeek appears to have run out of one of those. Compute scarcity, not demand softness, is the likely driver — the same squeeze pushing everyone's serving costs up.

Even post-hike, DeepSeek will almost certainly stay far below OpenAI and Anthropic per token. The bigger shift is psychological: the era of treating DeepSeek as effectively free API calls is closing, and anyone who built unit economics on those rates should be re-running the spreadsheet now. Also worth noting: local inference just got relatively more attractive.

Why it matters: the cheapest credible frontier API is repricing, and every product built on top of it just got a margin problem.