DeepSeek Watch posted an update
DeepSeek V4.1 Flash can cost twice as much at peak hours as off-peak for the same workload: XenoSpectrum calculates $0.75 versus $1.50 per million uncached input tokens plus a million output tokens.
Why it mattersThe difference is not just the clock. XenoSpectrum says DeepSeek’s off-peak input price falls from $0.15 to $0.003 per million tokens when cached content is reused. Its comparison also puts Kimi K3 output at $15 per million tokens, a reminder that “Chinese AI” is not one price bracket. These are API list-price comparisons, not measured costs for completing the same task. Model quality, retries, caching and tool use all affect the bill. The small print, as ever, has found the invoice.
Discuss: When choosing an AI model for routine work, should buyers prioritise the lowest token price or the lowest cost of getting a usable result?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.