Guides · dev · openrouter deepseek pricing
OpenRouter versus direct: where the Chinese model is actually cheaper
OpenRouter is a switchboard. The price on the model card is not a promise that you are buying from the lab, and it is not always the cheapest total once output tokens and cache rules are included.
Three patterns we checked
DeepSeek flash. Direct peak cache-miss is $0.30 / $1.20. Off-peak is $0.15 / $0.60. The OpenRouter card for V4.1 Flash matches peak, not off-peak. If your traffic can move to nights, weekends, or Chinese public holidays, direct is cheaper and the data path is DeepSeek itself. If you want one key for many labs and you can tolerate peak pricing, OpenRouter is the convenience.
DeepSeek V4 Pro. Direct peak is $1.32 / $3.96 and off-peak is $0.66 / $1.98, for the V4-Pro-0813 column. The OpenRouter id’s list price is a third number. Do not average them.
GLM-5.3 and Kimi K3. The lab endpoint on OpenRouter matches the lab’s page. The card’s headline price does not.
GLM-5.3 Flash is the counterexample: Z.AI’s page and the OpenRouter card both say $0.15 / $0.50. The anomaly is the flagship, not the whole lab.
What you buy with the markup or the discount
OpenRouter sells a single account, failover across hosts, and a card payment that does not require a lab-specific cloud account. That is worth money if your alternative is a China-only console. It is not worth money if you thought you were paying Z.AI’s rate and you were paying a host with a different data-processing story.
Direct DeepSeek off-peak is the clearest “direct is cheaper” case we can cite. Direct Kimi is the clearest “the list price was a tease” case.
| Model | Blended | vs Opus | vs Sol |
|---|---|---|---|
| deepseek-v4.1-flash | $0.198 | 98% cheaper | 95% cheaper |
| deepseek-v4-pro | $1.19 | 85% cheaper | 70% cheaper |
| glm-5.3 | $1.26 | 84% cheaper | 69% cheaper |
| glm-5.3-flash | $0.291 | 96% cheaper | 93% cheaper |
| kimi-k3 | $3.13 | 61% cheaper | 22% cheaper |
A rule
Quote a price only with a host and a date. “Kimi is $0.40 input” is the OpenRouter card on this snapshot, and it is the wrong sentence if the reader will call Moonshot. “GLM is 84% cheaper” uses the list blend. The official blend is computed in the note above and is a smaller discount.
Related: the price map, hosts outside China.
Questions
Is OpenRouter cheaper than DeepSeek?
For V4.1 Flash, OpenRouter matches DeepSeek's peak price. DeepSeek off-peak is half. For V4 Pro, the OpenRouter list price is not DeepSeek's published peak or off-peak rate.
Why is GLM-5.3 so cheap on OpenRouter?
The model-level price tracks a low-cost host. Z.AI's own endpoint on the same feed matches the $1.40 / $4.40 docs.
Sources
- OpenRouter models API, accessed 2026-10-16
- OpenRouter GLM-5.3 endpoints, accessed 2026-10-16
- OpenRouter Kimi K3 endpoints, accessed 2026-10-16
- DeepSeek API pricing, accessed 2026-10-16
- Z.AI pricing, accessed 2026-10-16