Updated September 22, 2026: Use
deepseek-flashfor V4.1 Flash. Old Flash aliases are documented as served by that version and billed at the Flash price. The updates page says V4 Pro API service continues after September 14, 2026, with billing unchanged. The pricing page listsdeepseek-v4-proas DeepSeek-V4-Pro-0813. Direct DeepSeek API usage has separate billing. This site did not call the API.
The current model names
DeepSeek released V4.1 Flash on September 10. Its API identifier is deepseek-flash; this is the recommended name for new integrations. The legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp identifiers remain accepted and now route to V4.1 Flash.
On September 10, this page recorded deepseek-v4-pro as V4 Pro and a scheduled switch to V4.1 Flash on September 14 at 04:00 UTC. On September 22, the updates page says that switch is not the current plan, and the pricing page still lists a separate Pro version. Treat that as documentation, not as an authenticated check. Record the model metadata and date returned by each run.
Use it in MonkeyCode
MonkeyCode includes DeepSeek V4.1 Flash with free access under your current account terms. Start a task without setting up a separate DeepSeek API integration. Direct API requests to DeepSeek are a separate service and follow the provider’s pricing. A free MonkeyCode entry does not mean the official DeepSeek API is free.
Check compatibility before switching an agent
Start with one text request using the developer example. Then check one tool call, a multi-turn continuation and error handling independently. A compatible request format does not prove identical tool behavior or persistent conversation state. DeepSeek has a Responses API guide; its compatibility table explicitly marks some fields and tools as unsupported or ignored.
Budget from a completed change
This guide deliberately does not reuse July dollar prices, peak-hour multipliers or a Pro/Flash price ratio. Open the current pricing link from the official API documentation when budgeting; keep the currency, cache category, model and check date with the quote. Count retries, tool execution, compute and reviewer corrections in addition to token usage.
Use the first-task exercise to compare an accepted result on your own setup. A vendor benchmark and a small local exercise answer different questions; neither establishes production performance for your repository.