DeepSeek-V3
Open-weightLast verified: June 2026· DeepSeek DeepSeek-V3 pricing
Open-weight model with aggressive pricing. Off-peak discounts may apply — confirm with the provider.
- Input / 1M
- $0.27
- Output / 1M
- $1.10
- Context
- 128K tokens
- Provider
- DeepSeek
Pricing verified June 2026. Prices change frequently. Always confirm against the provider’s official pricing page before relying on these figures for budgeting. Official pricing →
When to choose DeepSeek-V3
The value row: near-frontier general quality at open-weight prices, $0.27 in and $1.10 out, inside a factor of two of the cheapest model in this table. Off-peak discounts can push the effective rate lower still, so the figure here behaves as a ceiling rather than a fixed price. Two things decide whether it belongs in your stack: whether quality holds on your own routes, which only evals answer, and where the data goes, which is a question to settle with the provider before the hosted API is in a design.
DeepSeek-V3 sits close enough to DeepSeek-R1 on price that the two are usually decided together. That comparison is written once, on the models table, rather than argued twice from opposite sides.
What DeepSeek-V3 costs in practice
A 50M/10M month costs $24.50 — about 1.5 times what Llama 3.3 70B costs at its representative hosted rate, and a fraction of the $160 the same traffic runs to on Mistral Large. Filling the 128K window costs about three and a half cents of input. At those numbers the per-token decision stops being the interesting one and two others take over: whether quality holds on your own routes, and whether an offshore hosted API is acceptable for the data you would be sending it. Neither is answered by the price column.
| Scenario | Input / mo | Output / mo | Est. cost / mo |
|---|---|---|---|
| Prototype | 2M | 0.5M | $1 |
| Growing product | 50M | 10M | $25 |
| At scale | 500M | 100M | $245 |
Plug in your own numbers with the cost calculator.
Where DeepSeek-V3 fits in a routing setup
The routing question for an open-weight row is which way the risk runs. Downward, Llama 3.3 70B does the reference month for $16 against $24.50 and can be self-hosted, which removes the offshore-API question altogether. Upward, a hosted mid-tier row catches whatever fails quality — Mistral Large at $160 for the same volumes keeps the work in the EU where that is the constraint. Roughly a ten-fold gap separates those two ends, which is why the eval result matters more here than the rate card.
Cheaper alternatives to DeepSeek-V3
The models immediately below DeepSeek-V3 on blended cost, nearest first. A swap only pays if quality holds on the routes you move — see the LLM cost optimization playbook for how to prove that before you move them.
Frequently asked questions
How much does DeepSeek-V3 cost?
DeepSeek-V3 bills $0.27 per million input tokens and $1.10 per million output tokens — $0.48 per million blended at a three-to-one input:output split. A workload of 50M input and 10M output tokens per month works out at about $25. Confirm current pricing with the provider.
What is DeepSeek-V3's context window?
DeepSeek-V3 has a 128K-token context window. One request that fills the whole window costs about $0.03 at the input rate, before the reply is billed.
Where does DeepSeek-V3 sit on price?
At $0.48 per million blended — three input tokens for every output token — DeepSeek-V3 is dearer than 2 of the 19 models tracked here and cheaper than 16. Output bills at 4.1x the input rate, and the nearest cheaper model on this table is Llama 3.3 70B at $0.30 blended.
Related concepts: context window, model routing, prompt caching, evals.