DeepSeek · DeepSeek V4
DeepSeek V4 (Flash and Pro)
DeepSeek's V4-generation API lineup contains a lower-cost multimodal Flash model, served as deepseek-flash, and a higher-priced text-only Pro model. Both have a documented 1,048,576-token context window; compare accepted results on the same task before choosing.
Start here
- What this helps you do
- Decide whether DeepSeek V4 (Flash and Pro) is the right model for a job you have, or one to leave alone.
- What you leave with
- A plain read of what DeepSeek V4 (Flash and Pro) is for, what to avoid giving it, and the facts behind that.
- What to check before trusting it
- Thinking mode is on by default, with a non-thinking option available per request.
Treat the price as an invitation to test, not a reason to route everything here. Score it on accepted work against your current model, pin the current model name (deepseek-flash or deepseek-v4-pro, not the retired deepseek-v4-flash id) and confirm the data terms before any regulated or sensitive task touches it.
Use it for
- High-volume or cost-sensitive tasks where token price dominates
- Long-context document and code work up to roughly a million tokens
- Comparing an aggressively cheap option against a frontier baseline on the same task
Do not assume a low token price means a low total cost, and do not route sensitive or regulated material to it before reading DeepSeek's own data-handling terms, which are not established in the cited pricing and release documentation.
Effort without the jargon
The task is high-volume or cost-sensitive and the multimodal Flash model passes your sampled quality and failure checks.
The higher-priced text-only Pro model produces enough additional accepted value on your matched task to justify its 4.4x cache-miss input and 3.3x output rates; price alone does not establish higher quality.
Advanced recordCost, limits, privacy and operations
deepseek-flash is cheap enough to route a lot of volume through, but DeepSeek has not published a portable data-retention rule in the cited sources, so read its terms before sending anything sensitive.
- API cost
- Officially disclosedPeak API rates per million tokens are US$0.30 cache-miss input, US$0.006 cache-hit input and US$1.20 output for deepseek-flash (DeepSeek-V4.1-Flash); deepseek-v4-pro (DeepSeek-V4-Pro-0813) costs US$1.32, US$0.044 and US$3.96 respectively. Pro is 4.4x Flash for cache-miss input and 3.3x for output; cached input has a different ratio. Off-peak rates are half of peak. Peak windows are weekdays 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays. All other hours, weekends and Chinese public holidays in full are off-peak. Legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp names route to deepseek-flash at Flash prices.
- Context window
- Officially disclosed1,048,576 tokens
- Maximum output
- Officially disclosed393,216 tokens (384K)
- Knowledge cutoff
- Not disclosedNot disclosed in the current cited pricing and model documentation
- Recorded inputs
- Text · Images (Flash only; Pro does not support vision)
- Recorded output
- Text · Structured JSON · Tool calls
Data handling is a product decision
Not disclosedThe current cited pricing and model documentation does not establish a portable retention or training rule for V4. Confirm the API and product terms for the exact service before sending sensitive material.
Operational constraints
- Thinking mode is on by default, with a non-thinking option available per request.
- Off-peak prices are half the peak rate. Peak is weekdays 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays; weekends and Chinese public holidays are off-peak in full.
- Retention and training terms are not established in the current cited pricing and model documentation; verify the exact service terms separately.
- The original deepseek-v4-flash-vision-exp experimental vision model has been retired along with deepseek-v4-flash; both legacy names now route to deepseek-flash (DeepSeek-V4.1-Flash-1017) rather than remaining separate options.
- Exact limits are from a static documented response example, not an authenticated API response or an inference test.
- The later updates documentation says Pro service continues with unchanged billing; do not treat the older planned post-14-September routing as current Pro retirement.
Record details
- Evidence readiness
- informational
- Guidance confidence
- moderate
- Evidence revision
- deepseek-v4@v3
Unknown means the cited official record does not disclose a safe value. It is not an estimate. Prices are provider list prices in USD where stated and can change before this record's review date.
Keep the evidence separate
DeepSeek's current V4-generation API lineup is deepseek-flash, serving DeepSeek-V4.1-Flash-1017, and deepseek-v4-pro, serving DeepSeek-V4-Pro-0813. Its static documented Models Example lists 1,048,576 context tokens and 393,216 maximum output tokens for both. Flash supports image input; Pro is text-only. Thinking is on by default with a non-thinking option, and JSON output and tool calls are supported. Legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp requests route to Flash at its price. At peak, Flash costs US$0.30 cache-miss input, US$0.006 cache-hit input and US$1.20 output per million tokens; Pro costs US$1.32, US$0.044 and US$3.96. Those are 4.4x cache-miss input and 3.3x output ratios, with a different cached-input ratio. Off-peak rates are half of peak.
No sufficiently useful independent note has been added yet. That absence is not filled with our guess.
No Human Bit result is claimed; use this as evaluation guidance only.
Not scheduled
No Human Bit result is claimed for this model yet.
Sources
Compare DeepSeek V4 (Flash and Pro) with another route for a job →