DeepSeek’s V3 release announcement listed API prices, from February 8, 2025, of $0.27 per 1 million cache-miss input tokens, $0.07 per 1 million cache-hit input tokens, and $1.10 per 1 million output tokens. Those are historical V3 rates, not prices for a currently listed model. (DeepSeek V3 release announcement)
For a separate, current-model budget reference, DeepSeek’s pricing page lists deepseek-flash, served by DeepSeek-V4.1-Flash-1017. As of November 10, 2026, its listed off-peak rates are $0.15 per 1M cache-miss input tokens, $0.003 per 1M cache-hit input tokens, and $0.60 per 1M output tokens. These successor-model prices are not V3 prices. (DeepSeek Models & Pricing)
DeepSeek V3 API pricing
DeepSeek’s official V3 announcement gives the following historical API prices, effective February 8, 2025:
| V3 token type | Historical price per 1M tokens |
|---|---|
| Input, cache miss | $0.27 |
| Input, cache hit | $0.07 |
| Output | $1.10 |
The announcement says V3 pricing matched V2 until February 8, then lists these rates. They are the rates stated in that announcement—not a current quote or a complete pricing history for every later V3-related model. (DeepSeek V3 release announcement)
What about V3.2 pricing and API status?
V3.2 is a different version, so V3’s rates should not be applied to it. DeepSeek’s V3.2 release announcement says the model was live on the API and followed the same usage pattern as V3.2-Exp, but it does not state exact per-token rates. The V3.2-Exp announcement mentions an API price reduction of more than 50%, without giving the resulting exact rates. I can’t establish an exact historical V3.2 API price from those announcements. (DeepSeek V3.2 release announcement; V3.2-Exp announcement)
The current official pricing table lists deepseek-flash and deepseek-v4-pro, not V3 or V3.2. That establishes which models the page currently prices; it does not by itself establish whether any older model name still routes to a particular version. Morph’s October 18, 2026 report says the legacy names deepseek-chat and deepseek-reasoner were deprecated on July 24, 2026. I could not confirm from the official sources cited here whether those names still route anywhere or what version they would serve. (DeepSeek Models & Pricing; Morph’s October 2026 report)
Hosted API pricing and downloadable model weights are different: having access to model weights does not establish that the same model remains available through a hosted API under the same model ID or price.
Current DeepSeek API prices for a separate successor option
If you are estimating spend for a new API integration rather than historical V3 usage, DeepSeek’s pricing page lists these rates for deepseek-flash (DeepSeek-V4.1-Flash-1017). Prices are in USD per 1 million tokens.
| Rate period | Input, cache miss | Input, cache hit | Output |
|---|---|---|---|
| Off-peak | $0.15 | $0.003 | $0.60 |
| Peak | $0.30 | $0.006 | $1.20 |
DeepSeek defines peak hours as 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. Other hours, including weekends and Chinese public holidays, are off-peak. The pricing page says off-peak rates are half of peak rates. A cache-hit rate applies only to input that qualifies as a cache hit, so use the cache-miss rate for a conservative estimate unless you have measured your own cache-hit share. DeepSeek notes that prices can change; check its pricing page before finalizing a budget. (DeepSeek Models & Pricing)
Example monthly budget for deepseek-flash
Assume 100 requests per day for 30 days, with 1,000 input tokens and 300 output tokens per request, and assume all input is billed as cache misses:
- Monthly input: 100 × 30 × 1,000 = 3 million tokens
- Monthly output: 100 × 30 × 300 = 0.9 million tokens
- Off-peak estimate: (3 × $0.15) + (0.9 × $0.60) = $0.99
- Peak estimate at twice the rates: $1.98
This is a budget estimate for the currently listed deepseek-flash, not for DeepSeek V3. Your actual bill depends on input and output token totals, qualifying cache hits, and when requests run. DeepSeek states that model usage is billed based on input and output token counts. (DeepSeek Models & Pricing)
Production API: model, endpoint, and limits
DeepSeek’s current quick-start documentation uses the OpenAI-compatible base URL https://api.deepseek.com, an API key, and the model ID deepseek-flash. Its example sends the key as a Bearer token. The current pricing page maps that ID to DeepSeek-V4.1-Flash-1017; the change log says the October 17 update left its API calling method and pricing unchanged. (DeepSeek quick start; DeepSeek change log)
For that listed Flash model, DeepSeek documents a 1-million-token context, a maximum output of 384,000 tokens, JSON output, tool calls, and vision support. These are documented capabilities and limits, not a guarantee of quality, latency, capacity, or reliability for your production workload. Test the model on your actual tasks and traffic before deployment. (DeepSeek Models & Pricing)
Bottom line
- Historical V3 API price: $0.27/M cache-miss input, $0.07/M cache-hit input, and $1.10/M output, as announced effective February 8, 2025.
- V3.2: DeepSeek’s release materials confirm API availability at launch but do not establish an exact historical per-token price. Do not infer one from V3 or V4 rates.
- Current separate budget reference:
deepseek-flashis listed at $0.15/M cache-miss input, $0.003/M cache-hit input, and $0.60/M output off-peak, with double rates during specified peak hours. This is successor pricing, not V3 pricing. - Production route: The documented current model ID is
deepseek-flash, usinghttps://api.deepseek.comand Bearer-key authentication. Validate it against your own workload before relying on it.
Sources
Introducing DeepSeek-V3 | DeepSeek API Docs
DeepSeek-V3.2 Release | DeepSeek API Docs
Introducing DeepSeek-V3.2-Exp | DeepSeek API Docs
Models & Pricing | DeepSeek API Docs
Your First API Call | DeepSeek API Docs