Fixstars offers DeepSeek's open model on GPU servers inside Japan
Japanese software firm Fixstars now runs open-weight models on domestic servers, starting with DeepSeek, giving Japanese teams a way to keep data in the country.
The short version
- Fixstars announced the service on November 5, working with a firm called 国際調達情報
- First model offered: DeepSeek-V4.1-Flash-1017, from China's DeepSeek
- Input data is processed only on Fixstars' own GPU servers in Japan
- Fixstars says it does not save input or output data on its servers
- More models are planned; the service is reached through the FastMetal API
Why it matters
Many organizations want to use strong open models but cannot send documents to servers abroad. A hosted option inside Japan lets them try a DeepSeek model without buying GPUs or moving data overseas. The storage and processing claims are the company's own, so teams with strict rules should still verify them.
What Fixstars launched
As itmedia.co.jp reported, Fixstars, a software development company in Tokyo's Minato ward, announced on November 5 that it has started a service for using open-weight models on GPU servers in Japan. Open-weight means the trained model files are published, so others can run them. The company is working with 国際調達情報, a Chiyoda-ward firm known for domain registration services.
The first model is DeepSeek-V4.1-Flash-1017, from the Chinese developer DeepSeek. The source does not explain the "1017" at the end of the name. It may be a version label, but that is a guess, so check before assuming it matches a model you downloaded yourself.
Who does what, and where data goes
Fixstars manages the GPU servers and the models, and tunes them to run well. Customers reach the models through the API of FastMetal, an AI gateway service run by 国際調達情報. An API is a connection that lets your software send requests to the model directly.
According to the report, input data is handled only on GPU servers that Fixstars owns in Japan. The company also says its own technology speeds up the AI processing, and that its servers do not save inputs or outputs. Those are the company's statements, not independent test results.
What it costs
Pricing is per million tokens, which are the small chunks of text a model reads and writes. For DeepSeek-V4.1-Flash-1017 the rates are 48円 for input, 192円 for output and 0.96円 for cache reads. The source does not define cache reads. They usually mean the cheaper rate for input the service has already processed.
Fixstars plans to add more models later, though the source names none and gives no schedule.
What to watch
- Which models Fixstars adds after DeepSeek-V4.1-Flash-1017
- Whether the no-storage and Japan-only processing claims get independent verification
Try it
If your team has data-residency rules, ask the provider which servers handle requests and how long data is kept, and get the answers in writing.
Sources and links
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by itmedia.co.jp · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.