Pera Computing

API docs

Base URL and keys

Base URL https://api.peracomputing.com/v1, model kimi-k3.1-pro. Send your key as Authorization: Bearer <key> or x-api-key: <key>. For a key, see Credits or write to hello@peracomputing.com.

Endpoints

`POST /v1/chat/completions`	OpenAI format, with streaming, tool calls, logprobs and reasoning returned as `reasoning_content`.
`POST /v1/messages`	Anthropic format, with streaming, tools, images and reasoning returned as `thinking` blocks. `/v1/messages/count_tokens` is not offered.
`GET /v1/models`	The model list with context length and prices. No key needed.
`GET /v1/usage`	Your key's prices, credit, spend, remaining balance, and today's and 30-day totals.

OpenAI SDK

from openai import OpenAI

client = OpenAI(base_url="https://api.peracomputing.com/v1", api_key="sk-pe-...")
r = client.chat.completions.create(
    model="kimi-k3.1-pro",
    messages=[{"role": "user", "content": "Write a haiku about pears."}],
)
print(r.choices[0].message.content, r.usage)

Anthropic SDK

The Anthropic SDK adds /v1/messages itself, so its base URL has no /v1.

from anthropic import Anthropic

client = Anthropic(base_url="https://api.peracomputing.com", api_key="sk-pe-...")
m = client.messages.create(
    model="kimi-k3.1-pro",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Write a haiku about pears."}],
)
print(m.content, m.usage)

Reasoning and reasoning_effort

Reasoning arrives separately from the answer: as reasoning_content on /v1/chat/completions, as thinking blocks on /v1/messages. Reasoning tokens are output tokens and are billed as output.

Kimi K3.1 Pro always reasons. Its model guide lists reasoning_effort values low, high and max, with max as the default; set it at the top level of the request.

Usage reporting

Every non-streamed response carries usage. On /v1/chat/completions:

`prompt_tokens`	Input tokens, cached ones included
`prompt_tokens_details.cached_tokens`	Input tokens served from the prompt cache, billed at the cached-input price
`completion_tokens`	Output tokens, reasoning included
`reasoning_tokens`	The part of `completion_tokens` spent on reasoning
`total_tokens`	Prompt plus completion

For streams, send "stream_options": {"include_usage": true}; the last chunk before [DONE] then carries usage. On /v1/messages, usage comes as input_tokens, cache_read_input_tokens and output_tokens; in streams the final counts are in message_delta.

A request costs uncached input × input price + cached input × cached price + output × output price; the prices are on the home page. Caching is automatic: repeat the start of a conversation and its prefix is read from the cache. We keep token counts, never prompts or completions.

Errors

400, 413 and 422 mean the request itself needs fixing. 429 and 503 mean capacity is full: retry after the number of seconds in Retry-After; their body uses the error format of the endpoint you called. Served requests carry an X-Request-Id header; include it when you write to us about a request.