Last verified
JAPANESE-SPECIALISEDSAME CARD AS ITS BASE MODEL256K CONTEXTTOOLS BILL SEPARATELYNO SUBSCRIPTION TIER

Sakana Namazu API Pricing

Sakana Namazu is the lab's first in-house single model — everything else it sells, Fugu Ultra included, orchestrates other labs' frontier models. It lists at $0.95/M input, $4.00/M output and $0.15/M cached input, which is the striking part: Sakana states Namazu is built on Kimi K2.6, and Moonshot sells that base model at $0.95 / $4.00 too. You pay the base model's price for the Japanese post-training. What is not in the per-token rate is the tooling: web search bills at $7.00 per 1,000 calls and code execution at $0.12 per session-hour, on top of tokens.

Input - per 1M tokens
$0.95/M
Base token price standard
Output - per 1M tokens
$4.00/M
Thinking billed as output standard
Cached input
$0.15/M
Cache hit vs miss -84%
Effective - agentic blend
$0.59/M
92/8 split - 82% cache
§ 01 / TERMINAL

Run the numbers.

Calculator pre-loaded with Namazu's published rates. Tool fees are not in it — add $0.007 per search call and $0.12 per code-execution hour separately, and remember that tool results come back as billed input tokens on the next hop of the loop.

$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
Words equivalent (English)
Effective rate
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Real-world presets.

§ 03 / TOKENIZER

Paste text. See tokens. See cost.

Estimate · kimi-tokenizer-estimate · ≈3.6 chars/token Auto-counts as you type

This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.

Characters 549
Words 92
Tokens (estimated) 153 tokens
Cost as input · uncached $0.00015 USD
Cost as output · uncached $0.00061 USD
Cost as cached input $0.00002 USD
§ 04 / SHELF

Up against the shelf.

All models →
Model Input /M Output /M Effective blended Context Best for
Sakana Namazu $0.95 cache $0.15 $4.00 $0.59 current page 256K Japanese business and language work
Kimi K2.6 $0.95 cache $0.16 $4.00 $0.598 the base model 262K The model Namazu is built on
Fugu Ultra $5.00 cache $0.50 $30.00 $3.61 pricier 272K Sakana's multi-agent tier
Gemini 3.7 Flash $0.75 cache $0.075 $3.75 $0.481 cheaper 1M Wider context, no Japanese tuning
Claude Haiku 4.5 $1.00 cache $0.10 $5.00 $0.682 pricier 200K Cheap Western tier for the same work
GPT-5.6 Luna $0.20 cache $0.02 $1.20 $0.152 cheaper 1.05M Cheapest long-context alternative
§ 05 / DEEP LINKS

Specific scenarios.

All calculators →

Frequently asked.

What Namazu costs, what the tools add on top, and how it relates to the model underneath it.

Q · 01 How much does Sakana Namazu cost? +
$0.95/M input, $4.00/M output and $0.15/M on a cache hit, billed pay-as-you-go. Thinking tokens bill at the output rate. Under AI//COST's 92/8 agentic blend with 82% cache hits, the effective planning figure is $0.59/M.
Q · 02 Is the Japanese tuning charged for? +
No, and that is the notable thing about this card. Sakana states Namazu is built on Kimi K2.6, and Moonshot sells that base model at $0.95 / $4.00 — the same input and output rates. Namazu's cache hit is a cent cheaper ($0.15 against $0.16). You are paying the base model's price and getting the post-training for free, at least at list.
Q · 03 What do the built-in tools add? +
Web search: $7.00 per 1,000 calls, where a search query and a page-body fetch each count as one call — so about $0.007 per call, roughly half a planning turn's worth of tokens. Code execution: $0.12 per session-hour, counted from sandbox creation until the response completes. Both sit on top of tokens, and neither appears in the calculator above.
Q · 04 Why does a tool-using request cost more than the token count suggests? +
Because one API request becomes several internal model calls, and each re-sends the whole prompt. Sakana documents this plainly: search results, fetched page content and code output are billed as input tokens on the following internal call, and the usage object reports the total across all of them. The prompt prefix is normally served from cache at $0.15/M, which is what keeps a long loop from running away.
Q · 05 Is Namazu included in a subscription? +
No. Sakana's $20, $100 and $200 monthly tiers cover Fugu and Fugu Ultra; Namazu is pay-as-you-go only. If your usage pattern is what the subscription was bought for, Namazu is billed on top of it.
Q · 06 How does it differ from Fugu Ultra? +
Fugu Ultra is an orchestrator — it routes between one and three expert agents drawn from other labs' frontier models, and costs $5 / $30, about 6x Namazu's effective rate. Namazu is a single in-house model behind one endpoint. Different products with the same API: the only thing that changes between them is the model parameter.
Q · 07 What can it take as input? +
Text, images by URL or base64, and files — PDF, XLSX, CSV, DOCX and others up to 4 MB each. Context is 256K tokens, structured output is available through json_schema and json_object, and the endpoint speaks the OpenAI Responses and Chat Completions shapes as well as the Anthropic Messages shape.