DeepSeek API Model Pricing Catalog

Scrapes the DeepSeek API documentation site and returns one record per model with complete pricing, feature flags, context window sizes, and footnote annotations. Useful for cost-modelling batch inference workloads, comparing DeepSeek pricing against OpenAI/Anthropic, and tracking price changes over successive runs.

What it does

• Fetches api-docs.deepseek.com/quick_start/pricing — a static Docusaurus page, no browser required
• Parses the model comparison table and emits one dataset record per model column
• Extracts per-model: input price, cache-hit input price, output price, context window, max output, and feature flags
• Captures footnote text with dated pricing-change announcements (e.g. "cache-hit price reduced to 1/10 of launch price, effective 2026-04-26")
• Runs in under 20 seconds on default settings

Use cases

• Batch inference cost modelling — plug current prices into spreadsheets or cost-simulation tools
• Competitive benchmarking — compare DeepSeek pricing against OpenAI/Anthropic/Mistral
• Price-change monitoring — schedule daily runs and diff the dataset for changed prices
• Procurement decision support — CFO and engineering teams can track discount windows and cache savings

Input

Field	Type	Required	Description
maxItems	Integer	Yes	Maximum records to return. DeepSeek currently offers 2 models; set to 10 to capture all.

Example input

{
  "maxItems": 10
}

Output

One dataset record per model. All prices are in USD per 1 million tokens.

Field	Type	Description
model_name	String	API model identifier (e.g. deepseek-v4-flash, deepseek-v4-pro)
model_version	String	Display version name (e.g. DeepSeek-V4-Flash)
context_window_tokens	Number	Maximum context length in tokens
max_output_tokens	Number	Maximum tokens the model can generate per request
input_price_usd_per_million	Number	Cache-miss input price (USD/1M tokens)
input_price_cache_hit_usd_per_million	Number	Cache-hit input price — typically 1/10 of the cache-miss price
output_price_usd_per_million	Number	Output token price (USD/1M tokens)
off_peak_input_price_usd_per_million	Number	Off-peak input price, when active (null if no off-peak window currently offered)
off_peak_output_price_usd_per_million	Number	Off-peak output price, when active
off_peak_window_utc	String	Off-peak time window in UTC (e.g. 16:30-00:30 UTC), or null
currency_native	String	Denomination of the listed prices (USD)
effective_date	String	Date of the most recent pricing change noted in footnotes (ISO 8601)
deprecated	Boolean	Whether the model itself is marked deprecated
replacement_model	String	Replacement model name if deprecated, otherwise null
features	Array	API features enabled for this model (e.g. json_output, tool_calls, fim_completion)
notes	String	Full footnote and disclaimer text from the pricing page
source_url	String	URL scraped
source	String	Source domain (api-docs.deepseek.com)
last_changed	String	ISO date when a price change was detected vs. prior run, or null

Example output record

{
  "model_name": "deepseek-v4-flash",
  "model_version": "DeepSeek-V4-Flash",
  "context_window_tokens": 1000000,
  "max_output_tokens": 384000,
  "input_price_usd_per_million": 0.14,
  "input_price_cache_hit_usd_per_million": 0.0028,
  "output_price_usd_per_million": 0.28,
  "off_peak_input_price_usd_per_million": null,
  "off_peak_output_price_usd_per_million": null,
  "off_peak_window_utc": null,
  "currency_native": "USD",
  "effective_date": "2026-04-26",
  "deprecated": false,
  "replacement_model": null,
  "features": ["json_output", "tool_calls", "chat_prefix_completion", "fim_completion"],
  "notes": "The prices listed below are in units of per 1M tokens...",
  "source_url": "https://api-docs.deepseek.com/quick_start/pricing",
  "source": "api-docs.deepseek.com",
  "last_changed": null
}

Pricing

This actor charges $0.10 per start + $0.001 per record returned. A standard run with maxItems: 10 returns 2 records (one per active model) and costs approximately $0.102.

Notes

• DeepSeek updates their pricing page directly when pricing changes. The effective_date field reflects the most recent dated footnote on the page.
• The off_peak_* fields are reserved for DeepSeek's discount window feature (historically 50% off during 16:30–00:30 UTC). These fields return null if the feature is not currently listed on the page.
• Legacy model aliases deepseek-chat and deepseek-reasoner are being deprecated; they currently map to the non-thinking and thinking modes of deepseek-v4-flash respectively.