​

Usage object

OpenAI usage shape plus one field: cached_prompt_tokens.

​

Shape

{
  "prompt_tokens": 1500,
  "completion_tokens": 300,
  "total_tokens": 1800,
  "cached_prompt_tokens": 1200
}

​

cached_prompt_tokens

• You sent X-Malibu-Conversation and the request routed to the same provider as a recent turn on the same tag.
• The shared prefix between the two turns exceeds the provider’s cache-miss threshold (typically a few hundred tokens).

• No X-Malibu-Conversation header on the request.
• The sticky provider was unavailable and the request fell through to normal routing.
• The provider serving the request is legacy (pre prefix-cache billing) — safe to consume unconditionally.

​

Billing implication

​

Reading from the SDK

resp = client.chat.completions.create(
    model="mlx-community/Qwen2.5-7B-Instruct-4bit",
    messages=[...],
)

usage = resp.usage
print(usage.prompt_tokens, usage.completion_tokens, usage.total_tokens)
print("cached:", getattr(usage, "cached_prompt_tokens", 0))

​

Aggregated usage

curl -s -H "Authorization: Bearer $MALIBU_API_KEY" \
  "https://api.malibu.tech/v1/usage?from=2026-06-01&to=2026-06-30" | jq .