Requests and usage
Three request shapes, one bill. Chat Completions for OpenAI clients, Anthropic Messages for Claude Code, Responses for Codex. Each returns the same four token counts in its own usage object.
input_tokens (new, charged), memory_write_tokens (written to Native Context Memory, charged once), memory_read_tokens (served from memory, free) and output_tokens (charged). On every shape, prompt_tokens = input_tokens + memory_read_tokens exactly, and cached_tokens is memory_read_tokens. The price card has the prices; cost = Σ class × price.POST /v1/chat/completions
The OpenAI shape, what OpenRouter and the OpenAI SDKs read. Set stream: true for server-sent events; the final chunk carries the full usage object, and the adaptiq block is sent even without stream_options.include_usage, because harness status lines depend on it.
curl https://api.adaptiqai.com/v1/chat/completions \
-H "Authorization: Bearer $ADAPTIQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "adaptiq-deepseek-v4.1-flash",
"messages": [
{"role": "system", "content": "You are a precise assistant."},
{"role": "user", "content": "Summarise the decision."}
]
}'"usage": {
"prompt_tokens": 312000, "completion_tokens": 1840, "total_tokens": 313840,
"prompt_tokens_details": {"cached_tokens": 310500},
"completion_tokens_details": {"reasoning_tokens": 1200},
"adaptiq": {"input_tokens": 1500, "memory_write_tokens": 1500,
"memory_read_tokens": 310500, "output_tokens": 1840,
"cost_usd": "0.004580", "price_card": "2026-11-17"}
}Core fields
| Field | Required | Description |
|---|---|---|
model | Yes | A model ID returned by /v1/models |
messages | Yes | OpenAI-compatible message array |
stream | No | Enable server-sent event streaming |
max_tokens | No | Maximum generated tokens for this request |
POST /v1/messages
The Anthropic Messages shape, what Claude Code reads. cache_creation_input_tokens is memory_write_tokens; cache_read_input_tokens is memory_read_tokens. Streaming: message_start.usage carries the input and cache fields, message_delta.usage the final output_tokens. cache_control on blocks is accepted and ignored: memory is always on.
curl https://api.adaptiqai.com/v1/messages \
-H "x-api-key: $ADAPTIQ_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "Content-Type: application/json" \
-d '{
"model": "adaptiq-deepseek-v4.1-flash",
"max_tokens": 1024,
"system": "You are a precise assistant.",
"messages": [
{"role": "user", "content": "Summarise the decision."}
]
}'"usage": {"input_tokens": 1500, "cache_creation_input_tokens": 1500,
"cache_read_input_tokens": 310500, "output_tokens": 1840}POST /v1/responses
The Responses shape, what Codex reads. The same adaptiq block as in Chat Completions is sent as a sibling of usage on the final response.completed event.
curl https://api.adaptiqai.com/v1/responses \
-H "Authorization: Bearer $ADAPTIQ_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "adaptiq-deepseek-v4.1-flash",
"input": [
{"role": "developer", "content": "You are a precise assistant."},
{"role": "user", "content": "Summarise the decision."}
]
}'"usage": {"input_tokens": 312000, "input_tokens_details": {"cached_tokens": 310500},
"output_tokens": 1840, "output_tokens_details": {"reasoning_tokens": 1200},
"total_tokens": 313840}Headers on every response
All three shapes, streaming included, sent before the body. Rate-limit headers use the OpenAI names and the Anthropic unified names, so both Claude Code's and Codex's status lines show them.
x-adaptiq-request-id: req_...
x-adaptiq-price-card: 2026-11-17
x-adaptiq-cost-usd: 0.004580 # after settlement, where the protocol allows
x-adaptiq-balance-usd: 148.21 # otherwise on the next request
x-ratelimit-limit-requests, x-ratelimit-remaining-requests
x-ratelimit-limit-tokens, x-ratelimit-remaining-tokens, x-ratelimit-reset-tokens
anthropic-ratelimit-unified-status, anthropic-ratelimit-unified-reset,
anthropic-ratelimit-unified-representative-claim
retry-after: <= 60 # on 429 and 503