Sakana Fugu Ultra

Sakana models use canonical model IDs under sakana/.... In this phase, only sakana/fugu-ultra is public in the catalog. sakana/fugu is not public because its agent pool can vary and its price is not fixed.

Fugu Ultra is Sakana AI’s OpenAI-compatible multi-agent reasoning model. It is designed for hard coding, code review, scientific research, security analysis, and long-context synthesis tasks.

Endpoints

sakana/fugu-ultra works through two OpenAI-compatible endpoints:

Endpoint	Use
POST /v1/chat/completions	Chat Completions-compatible clients
POST /v1/responses	Modern OpenAI Responses-compatible clients

Chat Completions:

curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sakana/fugu-ultra:high",
    "messages": [
      { "role": "user", "content": "Review this TypeScript service for concurrency risks." }
    ],
    "max_tokens": 1200
  }'

Responses:

curl "$LLMTR_BASE_URL/v1/responses" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sakana/fugu-ultra:max",
    "input": "Analyze a technical decision across cost, risk, and maintenance.",
    "max_output_tokens": 1200
  }'

Reasoning

Fugu Ultra accepts only two reasoning effort levels:

Level	Alias	Notes
high	:high	Deep reasoning
xhigh	:xhigh, :max	Highest level; max is a compatibility alias

You can also send it in the body:

{
  "model": "sakana/fugu-ultra",
  "messages": [{ "role": "user", "content": "Review this architecture." }],
  "reasoning": { "effort": "xhigh" }
}

low, medium, minimal, and none are not valid for this model.

Image Input

Fugu Ultra supports text and image input. For Chat Completions, use an OpenAI-compatible image_url content part:

{
  "model": "sakana/fugu-ultra:high",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "Analyze the issue in this screenshot." },
        {
          "type": "image_url",
          "image_url": { "url": "data:image/png;base64,<base64>" }
        }
      ]
    }
  ]
}

Tool/function calling is not advertised as a public capability in this phase. The catalog will not show function_calling until live tool-call behavior is verified separately.

Pricing

Prices are USD per 1M tokens. Above 272K input context, Sakana’s long-context prices apply.

Metric	Up to 272K	Context above 272K
Input	$5	$10
Output	$30	$45
Cached input	$0.50	$1.00

When upstream returns prompt_tokens_details.cached_tokens, the gateway splits cached input into a separate CACHE_READ line. Fugu Ultra is processed by Sakana AI’s fixed agent pool and is not labeled as a Turkey-hosted LLMTR model.