# Preferred Networks (PFN)

> Preferred Networks, Inc. is a Tokyo-based AI company founded in March 2014 that vertically
> integrates the AI value chain from its own MN-Core AI processors and supercomputers through to
> generative-AI foundation models. Its commercial API product is the PLaMo API, a cloud service for
> the PLaMo family of Japanese large language models with an OpenAI-compatible interface.

Generated by API Evangelist from public Preferred Networks sources on 2026-08-26. Preferred Networks
does not publish an llms.txt of its own — https://www.preferred.jp/llms.txt returns the site's
Next.js 404 page and https://docs.plamo.preferredai.jp/llms.txt returns HTTP 404. This file is an
independent third-party summary, not a PFN publication.

## API

- [PLaMo API reference](https://docs.plamo.preferredai.jp/en/api): Full reference for the three
  published endpoints and every request/response object.
- [Platform overview](https://docs.plamo.preferredai.jp/en/): PLaMo API and PLaMo Chat.
- [Quickstart](https://docs.plamo.preferredai.jp/en/getting-started): Obtain a key, call the API with
  curl, openai or langchain-openai.
- [Console manual](https://docs.plamo.preferredai.jp/en/console): Tenant, project, role and API key
  model; billing limits and alerts.
- [Limitations](https://docs.plamo.preferredai.jp/en/limit): Published rate limits and tenant quotas.
- [Plans and pricing](https://plamo.preferredai.jp/api): Free, Standard, Provider and On-Premise.
- [Sign up](https://plamo.preferredai.jp/signup)
- [API console](https://console.platform.preferredai.jp/)
- [Terms of use](https://plamo.preferredai.jp/info/terms)

## Base URL and authentication

- Base URL: https://api.platform.preferredai.jp/v1
- Legacy, deprecated base URL: https://platform.preferredai.jp/api/completion/v1
- Auth: `Authorization: Bearer ${PLAMO_API_KEY}`. OpenAI/LangChain clients read `OPENAI_API_KEY`.
- A keyless request returns HTTP 400 `{"message":"missing key in request header"}` — not a 401 with a
  WWW-Authenticate challenge.

## Operations

- `POST /v1/chat/completions` — chat completion. Supports streaming (SSE, terminated by
  `data: [DONE]`), function calling (`tools`, `tool_choice`), JSON-Schema structured outputs
  (`response_format`), and reasoning controls (`reasoning`, `reasoning_effort` — `none` or `medium`).
  `n` accepts only 1 or 2. `max_tokens` defaults to 4096 and the request errors when
  input tokens + max_tokens exceeds the model context length.
- `POST /v1/tokenize` — token count and token sequence for a prompt or message list; also returns
  `max_model_len`. Useful for pricing an input before spending on a generation.
- `GET /v1/models` — list available models.
- `GET /v1/models/{model}` — retrieve one model.

## Models

- `plamo-3.0-prime` — 262,144-token context, 20,000 max output tokens, reasoning. Current.
- `plamo-3.0-prime-beta` — 65,536-token context, 20,000 max output tokens, reasoning.
  End of life 2026-07-31.
- `plamo-2.2-prime` — 32,768-token context, 4,096 max output tokens, no reasoning.
  End of life 2026-09-30; support merged into `plamo-3.0-prime`.

## Pricing (PFN-published, JPY per 1M tokens)

- Free — input 0, output 0. Announced but marked "in preparation" and not yet available.
- Standard — input 60, output 250 (unit price up to 128k tokens). Zero data retention, no training on
  submitted data, relaxed rate limits, email support.
- Provider — custom quote. Adds full reasoning body, per-customer guardrail policy, priority
  processing, dedicated support.
- On-Premise — quoted. Model weights, Docker image, appliance, Snowflake Marketplace / Sakura AI
  Engine / Amazon SageMaker, or the edge model PLaMo Lite.

## Rate limits

- 100 requests/minute per API key for the plamo-3.0-prime family.
- 1,000 requests/minute per API key for plamo-2.2-prime and earlier, and translation models.
- 1,000,000 JPY/month per tenant spend quota (increase on request).
- 10 API keys, 10 users, 50 projects per tenant (all increasable on request).
- No rate-limit response headers are documented.

## Agent surfaces

- MCP: `plamo-translate server` (from the first-party `plamo-translate` PyPI package) starts a local
  MCP server on http://localhost:8000/mcp for the PLaMo Translate model. This is a local install, not
  a hosted endpoint. The PLaMo API itself exposes no MCP server.
- A2A agent card: none. `/.well-known/agent-card.json` and `/.well-known/agent.json` return 404 on
  every PFN and PLaMo host.
- OpenAPI: none published. `/openapi.json`, `/openapi.yaml`, `/swagger.json`, `/v1/openapi.json`,
  `/api-docs`, `/docs` and `/redoc` all return 404 on the API host.
- Webhooks / AsyncAPI: none.

## Open source

- [GitHub — pfnet](https://github.com/pfnet)
- [PLaMo Translate CLI](https://github.com/pfnet/plamo-translate-cli) — Apache-2.0, PyPI
  `plamo-translate`.
- [PLaMo examples](https://github.com/pfnet-research/plamo-examples) — MIT, function-calling and
  usage examples.
- [Optuna](https://optuna.org/), [Chainer](https://chainer.org/),
  [PFRL](https://pypi.org/project/pfrl/), [pytorch-pfn-extras](https://github.com/pfnet/pytorch-pfn-extras),
  [pfio](https://github.com/pfnet/pfio), [pysen](https://github.com/pfnet/pysen).

## Company

- [Preferred Networks](https://www.preferred.jp/en/)
- [Generative AI foundation models](https://www.preferred.jp/en/business/genai/)
- [AI governance and AI policy](https://www.preferred.jp/en/company/aipolicy)
- [Engineering blog](https://tech.preferred.jp/en/blog/)
- [Privacy policy](https://www.preferred.jp/en/policy/)
- [Support](https://support.plamo.preferredai.jp/) — plamo-support@preferredai.jp