Short answer
The historical DeepSeek-V3 API rate card was $0.27 per million input tokens on a cache miss, $0.07 per million cache-hit input tokens, and $1.10 per million output tokens, in USD. These are historical rates—not a verified current price for a DeepSeek-V3 production API. DeepSeek’s current pricing page lists deepseek-flash and deepseek-v4-pro, not a model named DeepSeek-V3. (api-docs.deepseek.com)
Historical DeepSeek-V3 rates
| V3 token category | Historical price per 1 million tokens (USD) |
|---|---|
| Input, cache miss | $0.27 |
| Input, cache hit | $0.07 |
| Output | $1.10 |
DeepSeek’s original V3 announcement gives those three rates. A later pricing guide identifies them as December 2024 V3 rates and labels them historical reference only. Neither source establishes that you can still call a V3 API at those prices today. (api-docs.deepseek.com) (api-docs.deepseek.com)
Worked example: calculate the historical cost
For 1 million cache-miss input tokens and 1 million output tokens, the historical V3 arithmetic is:
(1 × $0.27) + (1 × $1.10) = $1.37
That is a historical illustration, not a current quote. For a workload with cache hits, count those input tokens separately and use the historical $0.07 rate only for that calculation. Do not assume the same cache pricing applies to another model or to a present-day endpoint.
A reusable calculation is:
Historical V3 cost = (cache-miss input tokens ÷ 1,000,000 × $0.27) + (cache-hit input tokens ÷ 1,000,000 × $0.07) + (output tokens ÷ 1,000,000 × $1.10).
What DeepSeek currently lists for API pricing
As of November 10, 2026, DeepSeek’s official pricing documentation lists two API models: deepseek-flash and deepseek-v4-pro. It identifies the served versions as DeepSeek-V4.1-Flash-1017 and DeepSeek-V4-Pro-0813. The official table does not list DeepSeek-V3 as a currently priced model. (api-docs.deepseek.com)
| Currently listed API model | Input, cache miss / 1M tokens | Input, cache hit / 1M tokens | Output / 1M tokens |
|---|---|---|---|
| deepseek-flash | $0.30 peak / $0.15 off-peak | $0.006 peak / $0.003 off-peak | $1.20 peak / $0.60 off-peak |
| deepseek-v4-pro | $1.32 peak / $0.66 off-peak | $0.044 peak / $0.022 off-peak | $3.96 peak / $1.98 off-peak |
These are current-model comparison figures, not V3 rates. DeepSeek says its off-peak prices are half of peak prices. Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays; all other hours—including weekends and Chinese public holidays—are off-peak. (api-docs.deepseek.com)
For example, 1 million cache-miss input tokens plus 1 million output tokens on deepseek-flash would total $1.50 at peak or $0.75 off-peak, before any other charges. That calculation uses the currently listed Flash rates; it says nothing about V3 pricing. (api-docs.deepseek.com)
Can you use DeepSeek-V3 in production at the historical rates?
The available official pricing page does not establish a current V3 rate or list a V3 model for API billing. So the responsible budgeting answer is not to treat $0.27/$0.07/$1.10 as an available production price. The official docs instead give current rates for deepseek-flash and deepseek-v4-pro. (api-docs.deepseek.com)
A captured third-party pricing article reports a scheduled retirement of the legacy names deepseek-chat and deepseek-reasoner after July 24, 2026. That report is not confirmation of V3-specific endpoint status, and those names should not be assumed to mean a currently available V3 endpoint. For production planning, use the exact model ID shown in DeepSeek’s current documentation and test that model against your own workload. (api-docs.deepseek.com)
The official current list supports a practical evaluation path: measure your workload on the listed model or models, then budget from the matching current rate card. The right choice depends on your quality, latency, and cost requirements; the historical V3 price alone cannot establish which current model will fit. (api-docs.deepseek.com)
How to build a reliable production budget
- Confirm the provider and exact model ID. DeepSeek-direct, a cloud marketplace, and a third-party host can have different rate cards. The figures above are DeepSeek-direct prices.
- Use that model’s current rate card. Keep cache-miss input, cache-hit input, and output as separate categories where the provider prices them separately.
- Estimate token volume for a representative period. Use measured or tested input and output tokens—not just the number of API calls. If caching applies, estimate the share of input tokens that actually qualify as cache hits.
- Apply any time-based pricing conditions. For the current DeepSeek V4-family schedule shown above, distinguish peak from off-peak usage. Do not apply that schedule to V3.
- Add only confirmed extra charges or terms. Verify any provider-specific fees, limits, and service conditions against the chosen model’s documentation.
For a current model, the basic estimate is:
Cost = (cache-miss input tokens × cache-miss rate + cache-hit input tokens × cache-hit rate + output tokens × output rate) ÷ 1,000,000.
Use the rates that match the model, provider, and pricing period. If you cannot confirm a current V3 model ID and rate card, you can report the historical V3 arithmetic—but you cannot turn it into a dependable current production budget.
Sources
Models & Pricing — DeepSeek API Docs