What is Sakana Namazu?
Sakana Namazu is Sakana AI's Japanese-language model, opened as an API on 3 August 2026 after a test version ran inside the free Sakana Chat app from March. It is not a model trained from scratch: Sakana took Moonshot AI's open-weight Kimi K2.6 and post-trained it on its own Japanese data, aiming at Japanese language and business norms, fewer unnecessary refusals and less political bias in its answers.
It reads text, images and files and writes text, with web search and code execution built in. It works with OpenAI-compatible code by changing the base URL. Trackers list a 256K context and 65,536 output tokens, though Sakana's announcement does not state a context window. The weights are not released.
It is pay-as-you-go in US dollars, listed at $0.95 per million input tokens, $0.15 cached and $4.00 output, with web search at $7 per 1,000 calls and code execution at $0.12 an hour. Like Sakana's Fugu, it is not offered in the EU, EEA, UK or Switzerland.
How much does Sakana Namazu cost?
| Plan | Price | What you get |
|---|---|---|
| Sakana Namazu API | $0.95 in / $4.00 out per 1M tokens | Cached input $0.15; thinking tokens billed as output |
| Web search | $7.00 per 1,000 calls | |
| Code execution | $0.12 an hour |
Pay-as-you-go: $0.95 input / $4.00 output per 1M tokens, cached input $0.15; web search $7.00 per 1,000 calls; code execution $0.12 an hour. Free in Sakana Chat (Japan).
Is Sakana Namazu worth it?
Namazu is a localisation of someone else's model. Sakana says it keeps Kimi K2.6's maths, knowledge and coding performance while doing better on three Japanese tests: instruction following (JFBench, from Preferred Networks), Japanese-to-English translation of names and honorifics, and political neutrality, where its own FairPoliticsQA score rose from 34.1% to 56.3%. Two of those three tests are Sakana's own, and no outside group has checked them.
The practical value is convenience for Japanese businesses: a domestic vendor, built-in web search and code execution, and yen invoicing on enterprise plans. Sakana frames it as a way to use a strong open model without the data and bias worries of using a foreign model as it comes.
The weakness is value. Kimi K2.6 is already a generation behind: Moonshot has since shipped Kimi K3 and K3.1 Pro. At $0.95/$4.00 per million tokens, Namazu costs more than MiMo-V2.6-Pro ($0.435/$0.87) and Step 5 Preview ($1.00/$2.70) on output, both of which score higher on independent tests and handle Japanese.
Sakana's announcement says nothing about how API data is handled, and its sister product Fugu trains on customer data unless you opt out. Japanese teams that need a local vendor should ask for the data terms in writing.
Where it earns its place
A Japanese company drafting customer-facing text
Write and check business Japanese with the right honorifics
Namazu is tuned on Sakana's Japanese data for exactly this. 10,000 emails of 1,500 input and 400 output tokens cost about $30.
A Japanese media or research team
Answer questions on current topics with less one-sided framing
Use the built-in web search ($7 per 1,000 calls) and Sakana's neutrality tuning, which raised its own FairPoliticsQA score from 34.1% to 56.3%.
A developer already using an OpenAI-style client
Try a Japanese-tuned model without changing code
Point the client's base URL at Sakana's API with a Sakana key; function calling and image input work as in OpenAI's format.
Before you commit
- Setup
- Create a key in Sakana's API console and set it as the base URL in any OpenAI-compatible client. Consumers in Japan can use it free in Sakana Chat.
- Integrations
- OpenAI-compatible API with function calling, image input, built-in web search and code execution; OpenRouter.
- Data and privacy
- Sakana's announcement does not say whether API data is retained or used for training; its Fugu product trains on customer data unless you opt out. The service is closed to the EU, EEA, UK and Switzerland pending GDPR work.
- Cost at scale
- $0.95/$4.00 per million tokens with $0.15 cached input. A million short requests of 2,000 input and 500 output tokens cost about $3,900; resellers list it from $0.76/$3.20.
- Limits
- Context listed by trackers as 256K with 65,536 output tokens; Sakana has not stated it. Closed weights.
What's good
- Tuned for Japanese instruction following, translation and business norms
- Web search and code execution built in
- OpenAI-compatible; switch by changing the base URL
- Japanese vendor with yen invoicing on enterprise plans
What's not
- Built on Kimi K2.6, a generation behind Kimi K3
- Two of its three Japanese tests are Sakana's own
- Costs more on output than higher-scoring open models
- Data handling not stated; not offered in the EU, EEA, UK or Switzerland
Best for
Japanese companies that want a domestic vendor and a Japanese-tuned model with tools built in, more than the lowest price per token.
Sakana Namazu alternatives
Sakana Namazu: common questions
- Is Sakana Namazu worth it?
- A sensible Japanese-tuned Kimi K2.6 with tools built in, but it costs more than newer open models that already handle Japanese well. We rate it 3.6 out of 5.
- How much does Sakana Namazu cost?
- Pay-as-you-go: $0.95 input / $4.00 output per 1M tokens, cached input $0.15; web search $7.00 per 1,000 calls; code execution $0.12 an hour. Free in Sakana Chat (Japan).
- What are the downsides of Sakana Namazu?
- Built on Kimi K2.6, a generation behind Kimi K3; two of its three Japanese tests are Sakana's own; costs more on output than higher-scoring open models; data handling not stated; not offered in the EU, EEA, UK or Switzerland.
- Who is Sakana Namazu best for?
- Japanese companies that want a domestic vendor and a Japanese-tuned model with tools built in, more than the lowest price per token.
- How much does Sakana Namazu cost for a team or in production?
- $0.95/$4.00 per million tokens with $0.15 cached input. A million short requests of 2,000 input and 500 output tokens cost about $3,900; resellers list it from $0.76/$3.20.
- Is Sakana Namazu safe for company data?
- Sakana's announcement does not say whether API data is retained or used for training; its Fugu product trains on customer data unless you opt out. The service is closed to the EU, EEA, UK and Switzerland pending GDPR work.
- What are the best Sakana Namazu alternatives?
- The closest alternatives we have reviewed are Kimi K3.1 Pro, MiMo-V2.6-Pro, Sakana Fugu (Fugu, Max and Ultra v2) and Qwen3.8-Max.
Answers checked Oct 25, 2026.