PLaMo
PLaMo is Preferred Networks' from-scratch Japanese foundation model, sold as a pay-as-you-go OpenAI-compatible API alongside a web chat and on-premise deployments. The current release, PLaMo 3.0 Prime, adds a reasoning mode and a 256K-token context window.
Pros
- Genuinely self-serve for a Japanese lab model — you register on the console and start calling an OpenAI-compatible Chat Completions endpoint at 60 yen per million input tokens and 250 yen per million output tokens, roughly a fifth of frontier-model rates for Japanese work
- PLaMo 3.0 Prime carries a 256K-token context window, a reasoning mode that is a first for a fully from-scratch Japanese model, and tool-calling strong enough to sit behind an agent. The Standard plan contractually neither retains your inputs nor trains on them, new accounts receive credits worth 10 million tokens, the reference documentation has a complete English version, and the same model can be taken on-premise into a closed network, a VPC or an edge device via Docker, Snowflake Marketplace, Amazon SageMaker or the lightweight PLaMo Lite build.
Cons
- Everything outside the reference docs is Japanese-first — the marketing site, the plan comparison, the terms of service and email support all assume a Japanese-reading customer, and billing is in yen, so a non-Japanese team is working around the edges of a product not aimed at them. The advertised Free tier is still marked as in preparation rather than open, reasoning traces come back as summaries only unless you qualify for the sales-gated Provider plan, the flat rate above 128K tokens is a launch-period concession rather than a standing price, and there is only one model family here — no embeddings, no vision, and no reason to pick it over a frontier provider for English-language work.
Cost
Standard plan 60 yen per 1M input tokens and 250 yen per 1M output tokens for PLaMo 3.0 Prime up to 128K tokens, with usage above 128K charged at the same rate during the launch period; a Free plan is announced but not yet open; the Provider plan for AI service vendors is individually quoted, as is on-premise licensing. New registrations receive credits worth 10 million tokens, valid for one month. Billing is monthly pay-as-you-go in Japanese yen.
Verdict
Best for teams shipping Japanese-language products who want native-grade Japanese output at a fraction of frontier API cost, and especially for regulated Japanese buyers who need the same model on-premise or inside a closed network; skip it if your workload is English-first or you need an onboarding path that does not run through Japanese.
Visit PLaMo