DeepSeek released V4.1-Flash to API users on September 10, 2026 and introduced new pricing for its Flash tier. The company’s official API documentation says the model supports both thinking and non-thinking modes, a one-million-token context window and a maximum output of 384,000 tokens. Reuters described it as the smallest member of DeepSeek’s new architecture, while Chinese technology outlet IT Home reported the user notices and the pricing change that took effect the same day. DeepSeek’s claims that the model is better on performance, cost and speed are vendor statements. They have not yet been established across a broad set of independent evaluations.
How the API pricing works
DeepSeek’s current pricing page lists V4.1-Flash at $0.003 per million cache-hit input tokens, $0.15 per million cache-miss input tokens and $0.60 per million output tokens during off-peak periods. Peak prices are $0.006, $0.30 and $1.20 respectively. The company defines peak hours as 01:00–04:00 UTC and 06:00–10:00 UTC on weekdays. IT Home’s report of the local yuan-denominated schedule said the new rates started at noon Beijing time on September 10. Because the official documentation warns that prices may change, developers should verify the live pricing page before estimating production costs or committing workloads.
Legacy Flash names now point to V4.1
The primary API model name is “deepseek-flash.” DeepSeek says the legacy names “deepseek-v4-flash” and “deepseek-v4-flash-vision-exp” remain accepted, but requests using them are served by V4.1-Flash and billed at the Flash rate. The documentation lists support for JSON output, tool calls, the Responses API, an Anthropic-compatible API and vision. DeepSeek also says V4.1-Flash has comprehensively surpassed V4 Pro in performance, cost, generation speed and total completion time. That comparison comes from the company’s own testing. Results may differ across coding, reasoning, multilingual, multimodal and agent tasks, so the claim should not be treated as an independent benchmark.
V4 Pro transition moves to September 14
Early notices created an expectation that V4 Pro traffic would be redirected sooner. DeepSeek’s updated documentation and IT Home’s September 10 report now place the orderly retirement step at noon Beijing time on September 14, 2026. From that point until a future V4.1 Pro release, requests sent to “deepseek-v4-pro” are planned to route to V4.1-Flash and receive Flash pricing. Applications may keep using the same endpoint name, but the underlying model will change. Teams should therefore retest output formatting, tool behavior, latency, safety controls and quality thresholds before the transition. A cheaper bill does not remove the operational risk created by a model substitution.
The change is especially relevant for automations that depend on consistent model behavior. Even when an API name remains unchanged, response length, tool-call order or error patterns can shift. Organizations can reduce disruption by comparing the models on a limited share of traffic, monitoring failures and keeping a rollback path. The official page lists a concurrency limit of 2,500 for Flash and 500 for Pro, although real latency will still depend on demand and workload characteristics.
What it means for developers
Lower token prices could be significant for chat, coding and agent workloads that repeatedly submit large contexts. Price alone, however, is not enough for a production decision. Accuracy, security, data-handling terms, regional availability, concurrency limits and version stability must also be evaluated. Reuters and IT Home independently reported the September 10 launch, while DeepSeek’s API page supplies the technical limits, routing rules and current prices. The company’s performance claims should remain qualified until independent tests cover a wider range of real workloads. The next checkpoints are whether the V4 Pro routing change occurs on September 14 as documented and how V4.1-Flash performs in external benchmark and deployment reports.
