Preferred Networks launches a Japanese-native AI model at under half of OpenAI's price
Preferred Networks released a Japanese-language model trained exclusively on domestic data sources — no English translation in the pipeline — at less than half the cost of comparable OpenAI tiers. For Japanese enterprise workloads, it eliminates both the latency penalty of translation-based workflows and the data-sovereignty risk of routing Japanese text through foreign infrastructure.
Source: asia.nikkei.com ↗
a model optimized for the Japanese language that offers performance similar to comparable models from companies like OpenAI at less than half the price
Preferred Networks announcement
Why this matters
- → Eliminates translation overhead for Japanese enterprise workloads, reducing latency and cost.
- → Keeps sensitive Japanese text on domestic infrastructure, avoiding data-sovereignty risks.
- → Undercuts OpenAI pricing by 50%+ for comparable performance on local use cases.
Japan's AI sovereignty play