tools
Sakana Fugu
Sakana Fugu is an API that orchestrates multiple AI models to handle complex coding, research, reasoning, and security tasks.
In inglese
Sakana Fugu coordinates specialized AI agents through one OpenAI-compatible API. It offers Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber for different balances of latency, quality, cost, and cybersecurity focus.
It is used by developers, researchers, security teams, and enterprises for coding, code review, paper reproduction, research, vulnerability analysis, and agent workflows. Usage is paid through monthly subscriptions or token-based billing; Fugu Cyber requires access approval, and service is unavailable in the EU/EEA.
Features
- OpenAI-compatible API for Fugu, Fugu Ultra, Fugu Max, and Fugu Cyber
- Dynamically selects and coordinates specialized models for each task
- Supports coding, reasoning, research, and other multi-step workflows
- Lets users exclude selected providers or models from Fugu's pool
- Fugu Ultra coordinates a deeper agent pool for difficult reasoning tasks
- Fugu Max optimizes model selection for cost and performance
- Fugu Cyber focuses on security analysis, vulnerability research, and threat investigation
- Integrates with Codex CLI and Claude Code
Use cases
- Review code and identify bugs across large codebases
- Reproduce papers by implementing, training, and evaluating algorithms
- Investigate patents, literature, and other research materials
- Analyze vulnerabilities and conduct scoped security assessments
- Build chatbots and agent products through one model endpoint
- Run long-running coding, research, or reasoning workflows
Pros
Cons
Latest updates
- Sakana Chatをアップデート：最新モデルに刷新、メモリー機能を追加
- Introducing Fugu Max and Fugu Ultra v2: Orchestrating the Pareto Frontier (Fugu Ultra v2)
Released Fugu Max and Fugu Ultra v2, orchestrating open and specialized models across capability and cost.
Capabilities
- Searches the web — “Fixed rates regardless of context length. web_search and web_fetch are billed at $0.007 per call.” source
- Acts for you — “From one simple request, Sakana Fugu worked autonomously for nearly four hours — reading the paper, implementing, training, evaluating, and analyzing the gaps.” source
- Runs code — “the agent ran 123 experiments over ~14 hours on a single H100 GPU.” source
- Developer API — “Sakana Fugu is available through an OpenAI-compatible API.” source
Pricing
- Starting price
- $20/mo
- Prices checked
- 2026-09-25
Standard Plan
- $20 /month
- Lightweight daily usage
- For occasional API calls, small experiments, and personal workflows
- Baseline allowance
Pro Plan
- $100 /month
- Focused working sessions
- For regular coding, review, research, and analysis
- 10× Standard Plan usage
Max Plan
- $200 /month
- Heavy long-running workloads
- For power users running deeper, longer tasks
- 20× Standard Plan usage