DeepInfra

Route your DeepInfra API calls through AI SpendOps for automatic usage tracking and cost attribution.

Configuration

Setting	Value
Route	/v1/deepinfra/*
Upstream	https://api.deepinfra.com/v1/openai
Auth header	Authorization: Bearer ...
Streaming usage	Native. stream_options is not injected

SDK base URL

https://proxy.aispendops.com/v1/deepinfra/v1

Example

curl https://proxy.aispendops.com/v1/deepinfra/v1/chat/completions \
  -H "Authorization: Bearer your-deepinfra-key" \
  -H "X-ASO-API-Key: aso_k_yourkey.secret" \
  -H "Content-Type: application/json" \
  -d '{"model":"meta-llama/Llama-3.3-70B-Instruct","messages":[{"role":"user","content":"Hello"}]}'

Usage fields

Field	Description
prompt_tokens	Input tokens
completion_tokens	Output tokens

A large catalogue changes the cost problem

DeepInfra's appeal is breadth: a long list of open-weight models at very different price points, from small models costing a few cents per million tokens to large ones costing orders of magnitude more. That makes model choice, rather than request volume, the main driver of your bill.

Two AI SpendOps features matter more here than on a single-family provider:

Per-model cost breakdown. Because every event records the exact model string DeepInfra returned, you can see spend split by model rather than by provider. A single team quietly switching from an 8B to a 70B model is a large cost change that a provider-level total will hide completely.

Model allow-lists. You can restrict an ASO key to an approved set of models, so a change in application code cannot silently move traffic onto a more expensive model. See Restrictions.

If you call a model AI SpendOps does not yet have a rate for, the usage event is still recorded in full and the model is flagged for pricing. Usage is never dropped for being unrecognised, so token counts stay complete even when a cost figure has to be filled in slightly later.

What AI SpendOps records

Field	Populated
cache_read_tokens	No
cache_write_tokens	No
reasoning_tokens	Only where the model returns completion_tokens_details
provider_cost	No. Cost is calculated from token counts and the current DeepInfra rate card

Notes

• The proxy does not inject stream_options for DeepInfra. Streamed responses carry usage natively, and any response that completes without a usage block is recorded with usage_source: estimate. See Streaming.
• Model identifiers are Hugging Face style paths such as meta-llama/Llama-3.3-70B-Instruct, and allow-list entries must use the full path.
• Full field definitions are in the usage event field reference.