Skip to main content
OpenAI’s LLM models are known for advanced conversational abilities, creative writing, and broad general knowledge.
Versatile agents, customer support, brainstorming.
Credits charged per 1,000 tokens used for prompts + replies if you do not connect your own OpenAI API key / account.
OpenAI models can process each of the following files:
  • .pdf
  • .jpg
  • .jpeg
  • .png
  • .webp
  • .gif

​
Understanding pricing & implicit caching

OpenAI models automatically cache repeated input tokens (like agent instructions and tool definitions) in multi-turn conversations, charging them at 90% less than uncached tokens. This happens automatically with no code changes required.

​
Pricing example: GPT-4o

Token TypeCost per 1,000 tokens
Uncached input tokens0.63 credits
Cached input tokens0.063 credits
Output tokens5 credits
Example: An agent with 19,000 tokens of instructions makes two calls:
  • First call: 19,000 input (uncached) + 2,000 output = 21.97 credits
  • Second call: 6,000 input (uncached) + 13,000 input (cached) + 1,500 output = 12.10 credits
  • Without caching, the second call would cost 19.47 credits
  • Savings: 38% reduction on the second call

​
Why UI numbers may look different

The task credit breakdown displays uncached input tokens only. Cached tokens are charged separately at the lower rate but not shown in the breakdown. The total cost is accurate and includes both.
If you see “6,000 input tokens” in the UI, there may be additional cached tokens charged at 90% less. Your actual costs are lower than expected because of automatic caching.
No markup policy: Relevance AI passes through OpenAI’s exact pricing, including all caching discounts. We do not add any markup to Vendor Credits.
For technical details about OpenAI’s prompt caching implementation, see OpenAI’s Prompt Caching Documentation.
If you connect your own OpenAI API key, you’ll see the same caching behavior and cost savings directly in your OpenAI billing.