Why This AI Release Matters to Your Business
Artificial intelligence is moving beyond the idea of using one chatbot for every task. Sakana AI’s latest release, Fugu Max and Fugu Ultra v2, points towards a more practical approach: an AI system that decides which specialised model should handle each part of your work.
For a Malaysian SME, that distinction matters. You may need AI to read supplier documents, answer customer questions, update a spreadsheet, check a product catalogue, or help a developer fix an application. These tasks do not require the same level of reasoning. Using the most powerful model for everything can create unnecessary complexity, slower responses, and higher usage charges. Sakana’s Fugu approach is designed to match each request with a suitable model through one application programming interface (API), according to MarkTechPost’s report.
What Happened
Sakana AI launched Fugu Max and Fugu Ultra v2 as new additions to its Sakana Fugu family. Fugu is not a single foundation model. Instead, it is a learned orchestrator that routes work across a pool of different models. The company makes both releases available through an OpenAI-compatible hosted API, while the models do not have open weights for self-hosting, according to the source article.
Fugu Max is aimed at getting strong results with efficient model selection. It adds open-weight and specialised models, including NVIDIA Nemotron models through Sakana’s collaboration with NVIDIA. Fugu Ultra v2 is aimed at difficult, multi-step work such as autonomous research, visual reasoning, structured-data analysis, and full-stack software development. Both versions use an orchestration architecture that can assign different roles to AI agents, including Thinker, Worker, and Verifier, based on Sakana’s technical work described in the report.
The release follows a rapid product timetable. Fugu entered beta in April, reached general availability in June, and added Fugu-Cyber and a Claude Code interface in July, as reported by MarkTechPost. This suggests that orchestration is becoming an active area of AI development rather than a distant research concept.
Why This Matters for Malaysian SMEs
Your business probably has several workflows that look simple but contain multiple steps. Consider a local distributor receiving an email with a purchase order. An automated system could identify the customer, extract product codes, compare quantities against inventory, check delivery details, prepare a response, and send the record to your accounting or enterprise resource planning system. One AI model may not be ideal for every step. A routing layer could use a smaller model for text extraction and a stronger model only when an unusual instruction or discrepancy needs review.
The same approach could help a Malaysian online retailer manage product information. One agent could translate or refine descriptions in English, Bahasa Malaysia, or Mandarin; another could check whether claims comply with your internal guidelines; a third could compare the final content with catalogue data. You would still need approval controls, especially for product claims, refunds, customer complaints, and regulated information. However, the orchestration model could reduce the amount of repetitive checking done manually.
Service businesses can also apply the concept to customer support. A routine question about operating hours or delivery status could go to a lightweight model. A complaint involving a delayed order, damaged goods, or an important account could be routed to a stronger reasoning model or a human employee. This is more useful than treating every customer message as equally difficult.
“The practical lesson is not to ask which AI model is best for everything. Ask which model is appropriate for each business task, and how you will verify the result before it affects a customer.”
Key Points at a Glance
| Release | Main focus | Potential SME use |
|---|---|---|
| Fugu Max | Strong output with efficient model routing | Document processing, support triage, catalogue updates, routine automation |
| Fugu Ultra v2 | Complex reasoning and multi-step tasks | Research, software work, structured-data analysis, difficult operational cases |
| Shared architecture | Orchestration across multiple models | One integration for different workflow requirements |
| Deployment | Hosted OpenAI-compatible API | Faster pilot projects where your team already uses compatible tools |
The table summarises the product positioning described in the source article.
What You Should Check Before Adopting It
First, map the workflow rather than starting with the technology. Write down the trigger, the information the system receives, each decision it makes, the action it takes, and the point where a staff member must approve the result. For example, an invoice workflow may allow automatic data extraction but require human approval before posting an accounting entry.
Second, test with your own Malaysian business documents. Generic benchmark scores do not prove that a system will understand your supplier formats, abbreviations, product names, Bahasa Malaysia phrasing, tax-related terminology, or internal approval rules. Sakana reports strong results across several benchmarks, including Terminal Bench 2.1, GPQA Diamond, AutomationBench, Chartography, and DeepSWE, but those results are vendor-reported signals rather than a substitute for your own pilot, as noted by MarkTechPost.
Third, review data handling carefully. Fugu is offered as a hosted API and is not presented as an on-premise, self-hosted model in the source report. Before sending payroll records, customer identification details, confidential contracts, or proprietary recipes, you should examine the provider’s retention, security, access, and data-processing terms. You also need clear internal rules about which employees can submit sensitive information.
Finally, plan for failure. An orchestrator can choose the wrong model, misunderstand a request, repeat an incorrect answer, or fail when an external service changes. Keep logs, add confidence checks, set approval thresholds, and maintain a manual process for important transactions. Automation should make your team more consistent, not remove accountability.
The Bigger Picture
Sakana AI’s announcement reflects a wider shift from individual AI assistants towards coordinated AI systems. In this model, the important capability is not only the intelligence of one model. It is the ability to decompose a task, assign roles, select tools, verify intermediate results, and combine the output into a useful business action.
This could make advanced AI more accessible to SMEs because you may not need to build a separate integration for every model. A single compatible interface can simplify experimentation, while routing logic handles some of the technical decisions behind the scenes. It may also reduce dependence on one model provider, although an orchestration provider becomes another important dependency that you must evaluate.
For your business, the sensible next step is a controlled pilot. Choose one repetitive workflow with measurable results, such as extracting information from purchase orders or sorting customer enquiries. Compare the AI-assisted process with your current process, measure accuracy and turnaround time, and record every exception. If the pilot performs reliably, expand to the next workflow.
Fugu Max and Fugu Ultra v2 do not mean that every SME should immediately deploy multi-agent AI. They do show where business automation is heading: specialised models working together, with stronger systems reserved for genuinely difficult decisions. The winners will not be the businesses that automate the most tasks. They will be the businesses that design clear workflows, protect sensitive data, and use the right level of AI for each job.
Ready to Streamline Your Operations?
Technology moves fast. Your operations should keep up. AutoRunBiz builds AI systems that run your daily workflows — from WhatsApp order capture to accounting. Book a free 15-min ops audit →