DeepSeek is included as a direct hosted API provider. The two catalog entries use the current endpoint names, deepseek-flash and deepseek-v4-pro, rather than treating retired model aliases as additional products. Their model versions and official sources appear in the offer details above.
The catalog models text input and output tokens for language calls. DeepSeek documents tool calling, JSON output, thinking and non-thinking modes for both endpoints. Both have a one-million-token context window and a maximum output of 384,000 tokens. Flash also supports vision, but this calculator's text workload does not estimate image-input costs.
Search and page retrieval remain separate services in a stack. The calculator does not turn tool-calling support into an assumption that external search, scraping, or other tool usage is included in the language-model price.
Each offer uses the published peak-hour rate for input tokens that miss the cache, plus the peak output rate. The model names explicitly say “peak, uncached” so this assumption stays visible when comparing providers or opening a locked scenario.
DeepSeek also publishes lower off-peak and cache-hit prices. Those discounts are excluded here because the current form does not collect request timing or cache-hit proportions. This is a conservative token-cost baseline, not a prediction that every request will be billed at the peak rate. Check the official pricing page for the schedule and discounted paths before reconciling an invoice.
Select an offer to open the calculator with that model locked. Describe monthly activity, or use exact mode when you know the number of calls and tokens per call. Input and output tokens are charged separately, and the engine checks the combined context and output limits before returning a complete estimate.
The preset chat, support, coding, and document quantities are token assumptions. They do not establish task quality or guarantee that an application will use the same number of tokens. Adjust the workload to match measured requests, including the billable output reported by the API.
JSON output does not by itself establish support for a strict JSON schema. The catalog leaves that capability unknown rather than promising schema enforcement. Requiring structured output in the calculator therefore excludes these entries until that requirement is verified.
No independent task-quality score is added as part of this pricing entry. A known price can support a cost comparison while evidence about suitability remains incomplete. Use your own representative requests to evaluate behavior, and inspect the displayed verification date and sources when planning a budget. Taxes, infrastructure, retries, and development costs are outside the estimate.
