For years, AI progress has centered on scaling particular person basis fashions: bigger parameters, longer context home windows, stronger reasoning, and higher device use. Sakana AI\u2019s Fugu factors elsewhere, behaving like one mannequin from the surface whereas coordinating a number of professional brokers internally.<\/p>\n
A single API name can set off direct answering, specialist delegation, intermediate verification, and remaining synthesis, hiding orchestration complexity behind a standard LLM interface. On this article, a sensible information to Fugu\u2019s structure, variants, pricing, benchmarks, entry, code, checks, enterprise match, trade-offs, and use instances.<\/p>\n
What’s Sakana Fugu?\u00a0<\/h2>\n
Sakana Fugu is an OpenAI-compatible managed mannequin API that appears like a single LLM however works as a multi-agent system internally. Builders ship a immediate to 1 mannequin ID, corresponding to fugu or fugu-ultra, whereas Fugu handles agent choice, function task, coordination, verification, and remaining response.<\/p>\n
As a substitute of manually constructing planner, coder, reviewer, researcher, or supervisor brokers with frameworks like LangGraph, AutoGen, or CrewAI, groups get orchestration packaged into the mannequin itself. This reduces the necessity to handle prompts, routing, retries, reminiscence, state, monitoring, and failure restoration.<\/p>\n
Why the naming issues\u00a0<\/h3>\n
The identify \u201cSakana<\/span>\u201d means fish in Japanese. The corporate typically frames its analysis round collective intelligence, just like how a faculty of fish can behave as one coordinated system. Fugu follows that concept. Many brokers coordinate behind one interface.\u00a0<\/p>\n Most manufacturing AI methods as we speak fall into one in all three patterns:\u00a0<\/p>\n Single-model prompting\u00a0<\/p>\n Instrument-augmented LLM purposes\u00a0<\/p>\n Manually designed multi-agent workflows\u00a0<\/p>\n Single-model prompting is easy, however it will possibly fail on complicated duties that require planning, execution, verification, and iteration.\u00a0<\/p>\n Instrument-augmented LLMs enhance usefulness by connecting fashions to go looking, databases, code execution, APIs, or enterprise methods. However the mannequin nonetheless often acts because the central reasoning engine.\u00a0<\/p>\n Multi-agent workflows go additional. They divide work throughout specialised brokers. For instance:\u00a0<\/p>\n A planner breaks down the duty.\u00a0<\/p>\n A researcher gathers context.\u00a0<\/p>\n A coder writes code.\u00a0<\/p>\n A reviewer checks for correctness.\u00a0<\/p>\n A verifier checks the reply.\u00a0<\/p>\n A supervisor coordinates the method.\u00a0<\/p>\n This may enhance reliability on tough duties, however constructing it nicely is tough. Groups should reply many system design questions:\u00a0<\/p>\n Which agent ought to deal with which job?\u00a0<\/p>\n How ought to brokers talk?\u00a0<\/p>\n When ought to the system cease?\u00a0<\/p>\n How ought to intermediate outputs be verified?\u00a0<\/p>\n How ought to value and latency be managed?\u00a0<\/p>\n How ought to failures be recovered?\u00a0<\/p>\n How ought to compliance restrictions be utilized?\u00a0<\/p>\n Fugu makes an attempt to make this simpler by turning multi-agent orchestration right into a model-level functionality. The developer doesn’t have to design each agent interplay manually.\u00a0<\/p>\n Sakana Fugu was launched as Sakana AI\u2019s industrial multi-agent orchestration product. The preliminary beta positioned it as a system that coordinates swimming pools of frontier basis fashions for coding, arithmetic, scientific reasoning, analysis, and complicated evaluation.\u00a0<\/p>\n The newest Fugu launch makes the product simpler to entry via Sakana\u2019s console and an OpenAI-compatible API. The core launch message is easy: builders can plug multi-agent intelligence into present workflows with out rewriting their utility round a brand new SDK or orchestration framework.\u00a0<\/p>\n Sakana Fugu is available in two principal mannequin choices: Fugu and Fugu Extremely.\u00a0<\/p>\n Fugu is the default mannequin for on a regular basis work. It balances efficiency and latency. It’s appropriate for coding help, code assessment, chatbots, inside assistants, doc evaluation, and interactive workflows the place response time issues.\u00a0<\/p>\n A key level is that Fugu can path to the perfect mannequin primarily based on the duty. It additionally permits customers to choose particular brokers out of the mannequin pool, which can assist with information, privateness, compliance, or organizational necessities.\u00a0<\/p>\n Fugu Extremely is optimized for max reply high quality. It coordinates a deeper pool of professional brokers and is meant for onerous, high-stakes, multi-step issues. In response to the Sakana, Fugu Extremely can route between one to 3 brokers relying on the issue.\u00a0<\/p>\n Fugu Extremely is best suited to workloads the place accuracy, depth, and persistence matter greater than latency. Examples embrace:\u00a0<\/p>\n Paper replica\u00a0<\/p>\n Kaggle-style information science workflows\u00a0<\/p>\n Cybersecurity evaluation\u00a0<\/p>\n Literature assessment\u00a0<\/p>\n Patent investigation\u00a0<\/p>\n Deep technical analysis\u00a0<\/p>\n Complicated code assessment\u00a0<\/p>\n Scientific reasoning\u00a0<\/p>\n Function\u00a0 Greatest for\u00a0 Design aim\u00a0 Agent pool\u00a0 Latency\u00a0 Value\u00a0 Advisable customers\u00a0 Fundamental trade-off\u00a0 Fugu\u2019s structure could be understood as a managed orchestration layer wrapped inside a mannequin API.\u00a0<\/p>\n From the surface, the movement appears to be like like this:\u00a0<\/p>\n Internally, the system is nearer to this:\u00a0<\/p>\n Sakana Fugu exposes a single API whereas internally coordinating a pool of specialised fashions. The consumer sends one request, and Fugu handles routing, delegation, verification, and synthesis. \u00a0<\/p>\n The developer interacts with an ordinary API floor. This issues as a result of Fugu helps OpenAI-compatible endpoints, so groups can reuse present OpenAI SDK purchasers with a special base URL and API key.\u00a0<\/p>\n The orchestrator is the core intelligence layer. It decides how the duty must be dealt with. For less complicated duties, it might reply with minimal orchestration. For complicated duties, it will possibly coordinate a number of professional brokers.\u00a0<\/p>\n Fugu has entry to a pool of underlying fashions or brokers. These brokers might have totally different strengths throughout coding, reasoning, analysis, long-context evaluation, or different specialised duties.\u00a0<\/p>\n As a substitute of hardcoding a workflow, Fugu dynamically selects which agent or brokers to make use of. That is essential as a result of mannequin strengths are sometimes task-specific. One mannequin might carry out higher at code technology, one other at mathematical reasoning, one other at long-context synthesis.\u00a0<\/p>\n The orchestrator can break down a posh job into subtasks. It may ship centered directions to totally different brokers and management what context every agent receives.\u00a0<\/p>\n For tough duties, the system can use verification-style conduct. One agent might clear up, one other might critique or validate, and the orchestrator might mix the outcomes.\u00a0<\/p>\n The ultimate reply is returned as a single response. The consumer doesn’t see the total inside agent graph. .\u00a0<\/p>\n Fugu has two pricing modes: pay-as-you-go and subscription plans.\u00a0<\/p>\n Pay-as-you-go\u00a0<\/p>\n Pay-as-you-go is designed for heavier manufacturing workloads. Sakana says consumption-based tokens are served at increased precedence than monthly-plan tokens.\u00a0<\/p>\n Fugu pricing\u00a0<\/p>\n Fugu pricing relies on the energetic agent setup.\u00a0<\/p>\n Energetic brokers\u00a0 1 agent\u00a0 A number of brokers\u00a0 That is essential as a result of many multi-agent methods change into costly when every mannequin name is billed individually. Fugu\u2019s pricing mannequin tries to keep away from stacking mannequin charges throughout brokers.\u00a0<\/p>\n Fugu Extremely pricing\u00a0<\/p>\n Fugu Extremely has mounted pricing for fugu-ultra-20260615 per 1M tokens.\u00a0<\/p>\n Token kind\u00a0 Enter\u00a0 Output\u00a0 Cached enter\u00a0 Subscription plans\u00a0<\/p>\n Subscription plans are designed for people and on a regular basis hands-on use. Each tier consists of each Fugu and Fugu Extremely.\u00a0<\/p>\n Plan\u00a0 Normal\u00a0 Professional\u00a0 Max\u00a0 Sakana reviews Fugu and Fugu Extremely benchmark scores throughout coding, reasoning, science, agentic duties, long-context reasoning, and cybersecurity-style analysis.\u00a0<\/p>\n Sakana Fugu and Fugu Extremely in contrast with frontier baseline fashions throughout coding, reasoning, science, long-context, and agentic benchmarks. \u00a0<\/p>\n Benchmarks are helpful, however they shouldn’t be handled as direct manufacturing ensures. Fugu\u2019s benchmark profile suggests three sensible insights.\u00a0<\/p>\n 1. Fugu is strongest when duties require orchestration\u00a0<\/p>\n The strongest use case shouldn’t be a easy one-shot reply. The mannequin is designed for duties that profit from decomposition, professional choice, verification, and synthesis.\u00a0<\/p>\n Examples:\u00a0<\/p>\n Debug this repository.\u00a0<\/p>\n Assessment this pull request.\u00a0<\/p>\n Reproduce this analysis paper.\u00a0<\/p>\n Examine this patent panorama.\u00a0<\/p>\n Analyze a doable safety vulnerability.\u00a0<\/p>\n Examine a number of technical approaches and suggest one.\u00a0<\/p>\n 2. Extremely shouldn’t be all the time routinely higher\u00a0<\/p>\n Fugu Extremely is optimized for reply high quality, however Fugu can outperform it on some benchmarks. Builders ought to benchmark each fashions on their very own workload earlier than standardizing.\u00a0<\/p>\n A sensible routing technique might be:\u00a0<\/p>\n Use fugu for interactive work.Use fugu-ultra for complicated, high-value duties.Fallback to fugu when latency or value issues. \u00a0<\/p>\n 3. Multi-agent efficiency comes with hidden complexity\u00a0<\/p>\n Despite the fact that Fugu hides orchestration complexity from the developer, the underlying system nonetheless performs extra work. This may have an effect on latency, value, and observability.\u00a0<\/p>\n Groups ought to monitor:\u00a0<\/p>\n Whole tokens\u00a0<\/p>\n Orchestration tokens\u00a0<\/p>\n Latency by job kind\u00a0<\/p>\n High quality by workload class\u00a0<\/p>\n Failure instances\u00a0<\/p>\n Mannequin model conduct\u00a0<\/p>\n Value per profitable consequence\u00a0<\/p>\n Sakana fugu documentation: https:\/\/console.sakana.ai\/get-started<\/p>\n Go to the Sakana console API key web page login and create API:\u00a0https:\/\/console.sakana.ai\/api-keys<\/p>\n Create an API key and retailer it securely. The secret’s proven solely as soon as.\u00a0<\/p>\n export FUGU_API_KEY=”your_api_key_here”export FUGU_BASE_URL=”https:\/\/api.sakana.ai\/v1″ \u00a0<\/p>\n pip set up openai \u00a0<\/p>\n import os shopper = OpenAI( response = shopper.responses.create( print(response.output_text)<\/p>\n import os shopper = OpenAI( response = shopper.responses.create( print(response.output_text)<\/p>\n Sakana Fugu stands out as a result of it shifts the abstraction layer. As a substitute of providing simply one other massive mannequin, it packages multi-agent orchestration behind a mannequin API.<\/p>\n For builders, this implies simpler entry to agentic workflows with out constructing complicated orchestration methods from scratch. For technical leaders, it gives a managed approach to enhance reasoning, coding, analysis, and evaluation whereas lowering dependence on a single mannequin supplier.<\/p>\n Fugu is finest suited to complicated, ambiguous, high-value duties reasonably than easy chatbot prompts. Nonetheless, groups ought to undertake it fastidiously, given its restricted routing transparency, doable latency, unclear token accounting, and regional constraints.<\/p>\n The only approach to consider Fugu is that this: it’s not only a mannequin you immediate. It’s a mannequin that manages different fashions. That makes it an essential step towards the following technology of AI purposes.<\/p>\nWhy Multi-Agent System as a Mannequin Issues\u00a0<\/h2>\n
Sakana Fugu Launch Overview\u00a0<\/h2>\n
Fugu vs Fugu Extremely\u00a0<\/h2>\n
Fugu\u00a0<\/h3>\n
Fugu Extremely\u00a0<\/h3>\n
Comparability desk\u00a0<\/h3>\n
\n Fugu\u00a0
\n Fugu Extremely\u00a0<\/p>\n
\n On a regular basis coding, chat, assessment, interactive workflows\u00a0
\n Laborious reasoning, analysis, high-stakes evaluation\u00a0<\/p>\n
\n Stability high quality and latency\u00a0
\n Maximize high quality\u00a0<\/p>\n
\n Versatile, with opt-out help\u00a0
\n Fastened full pool\u00a0<\/p>\n
\n Decrease\u00a0
\n Greater\u00a0<\/p>\n
\n Is determined by energetic underlying agent tier\u00a0
\n Fastened token pricing\u00a0<\/p>\n
\n Builders, product groups, inside instruments\u00a0
\n Researchers, superior builders, enterprise evaluation groups\u00a0<\/p>\n
\n Much less depth than Extremely\u00a0
\n Greater value and response time\u00a0<\/p>\n<\/figure>\n<\/div>\nStructure: How Fugu Works Internally\u00a0<\/h2>\n
![\"flowchart\"](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2026\/06\/Gkld.webp\")
<\/figure>\n<\/div>\n![\"Internal](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2026\/06\/image-31.webp\")
<\/figure>\n<\/div>\nCore structure parts\u00a0<\/h2>\n
1. API gateway\u00a0<\/h4>\n
2. Orchestrator mannequin\u00a0<\/h4>\n
3. Agent pool\u00a0<\/h4>\n
4. Dynamic routing\u00a0<\/h4>\n
5. Delegation and communication\u00a0<\/h4>\n
6. Verification\u00a0<\/h4>\n
7. Synthesis\u00a0<\/h4>\n
Pricing \u00a0<\/h2>\n
\n Billing rule\u00a0<\/p>\n
\n Pay the usual price for the precise underlying mannequin\u00a0<\/p>\n
\n Charges will not be stacked. You might be charged one price primarily based on the top-tier mannequin concerned\u00a0<\/p>\n<\/figure>\n<\/div>\n
\n Normal value\u00a0
\n Context higher than 272K\u00a0<\/p>\n
\n $5 per 1M tokens\u00a0
\n $10 per 1M tokens\u00a0<\/p>\n
\n $30 per 1M tokens\u00a0
\n $45 per 1M tokens\u00a0<\/p>\n
\n $0.50 per 1M tokens\u00a0
\n $1.00 per 1M tokens\u00a0<\/p>\n<\/figure>\n<\/div>\n
\n Worth\u00a0
\n Greatest for\u00a0
\n Utilization\u00a0<\/p>\n
\n $20\/month\u00a0
\n Light-weight every day utilization, occasional API calls, small experiments\u00a0
\n Baseline allowance\u00a0<\/p>\n
\n $100\/month\u00a0
\n Common coding, assessment, analysis, and evaluation classes\u00a0
\n 10x Normal utilization\u00a0<\/p>\n
\n $200\/month\u00a0
\n Heavy long-running workloads\u00a0
\n 20x Normal utilization\u00a0<\/p>\n<\/figure>\n<\/div>\nBenchmark Outcomes\u00a0<\/h2>\n
Technical Palms-on: Utilizing Sakana Fugu API\u00a0<\/h2>\n
1: Create an API key\u00a0<\/h4>\n
![\"\"](\"https:\/\/cdn.analyticsvidhya.com\/wp-content\/uploads\/2026\/06\/image4-12-scaled.webp\")
<\/figure>\n<\/div>\n2: Set setting variables\u00a0<\/h4>\n
3: Set up the OpenAI Python SDK\u00a0<\/h4>\n
4: Primary Responses API name\u00a0<\/h4>\n
\nfrom openai import OpenAI<\/p>\n
\n api_key=os.environ[“FUGU_API_KEY”],
\n base_url=os.environ.get(“FUGU_BASE_URL”, “https:\/\/api.sakana.ai\/v1″),
\n)<\/p>\n
\n mannequin=”fugu”,
\n enter=”Clarify Sakana Fugu in easy phrases for a software program engineer.”,
\n)<\/p>\nStep 5: Use Fugu Extremely for tougher reasoning\u00a0<\/h4>\n
\nfrom openai import OpenAI<\/p>\n
\n api_key=os.environ[“FUGU_API_KEY”],
\n base_url=os.environ.get(“FUGU_BASE_URL”, “https:\/\/api.sakana.ai\/v1″),
\n)<\/p>\n
\n mannequin=”fugu-ultra”,
\n directions=”You’re a senior AI architect. Be exact and technical.”,
\n enter=”””
\nExamine single-agent LLM methods, manually designed multi-agent workflows,
\nand Sakana Fugu-style multi-agent methods as a mannequin.
\nGive attention to structure, value, latency, observability, and governance.
\n“””,
\n)<\/p>\nConclusion\u00a0<\/h2>\n
Ceaselessly Requested Questions<\/h2>\n
![\"Harsh](\"https:\/\/av-eks-lekhak.s3.amazonaws.com\/media\/lekhak-profile-images\/converted_image_0fBqNLi.webp\")
<\/p><\/div><\/div>\n