AI Artifacts: Diagrams, Prototypes and Mini-Apps
An AI can draw a diagram or code a small tool from a single request. What the 2026 benchmarks show, the useful cases for SMBs, and the limits you should know.
Ask an AI to "build me a discount simulator based on order volume" and get a small clickable tool a minute later: in 2026, that has become routine. OpenRouter now tests more than 200 models on their ability to draw, and about thirty on their ability to code a playable game from a single prompt. For an SMB, these "artifacts" open up an underrated use: building the small tools you are missing yourself.
At a glance
- An artifact is an object produced by AI: a diagram, dashboard, calculator, prototype or mini-game.
- In October 2026, drawing costs $0.000032 (Mistral Nemo) and coding a playable game $0.003 (gpt-oss-120b), according to OpenRouter.
- Ideal for prototyping and illustrating. Put guardrails in place as soon as a tool touches internal data or is used to make decisions.
Transparency: this article is published by GENIAL. The results cited come from the public OpenRouter Benchmarks page, accessed on October 14, 2026. For artifacts, OpenRouter publishes cost and speed, without a quality score.
What is an AI-generated artifact?
An artifact is a usable object that a text model creates by writing code: an SVG diagram, an HTML page, a small program. You do not see the code, you see the result: an org chart, a chart, a form, a calculator.
That is the difference between asking the AI "explain our ordering process to me" (a written answer) and "draw our ordering process" (a diagram you can paste into a presentation).
What do artifact benchmarks measure?
OpenRouter publishes two tests in which text models take on unusual tasks:
| Benchmark | What the model has to do | Models tested | Cheapest | Fastest |
|---|---|---|---|---|
| Sketch | Draw from a prompt; the result is scored as an image | 204 | Mistral Nemo, $0.000032 | Hy-MT2-1.8B (Tencent), 2 s |
| Games | Build a playable game from a single prompt, in one attempt | 29 | gpt-oss-120b (OpenAI), $0.003 | Gemini 3.5 Flash Lite (Google), 41 s |
Source: OpenRouter Benchmarks, October 14, 2026.
The Games test is the most telling for a business. A playable game requires logic, an interface and rules that work together on the first try. A model that pulls this off can also produce a calculator, a training quiz or a sales simulator.
Which artifacts are useful in an SMB?
| Artifact | Concrete example | Benefit |
|---|---|---|
| Process diagram | The path of an order, from request to invoicing | An hour of formatting saved |
| Calculator | Discount simulator, lead time estimate, profitability calculation | Sales reps answer without waiting for the sales administration team |
| Training quiz | Interactive questionnaire on a safety procedure | More engaging training, measurable results |
| Screen mockup | The future time-off request form | A specification everyone understands |
| Dashboard | Visualization of an Excel export | Trends you can read at a glance |
The most profitable case is often the mockup: showing a clickable prototype to a vendor or to management saves weeks of back-and-forth over a written specification.
What limits should you know before relying on them?
- The artifact looks right, but it is not necessarily right. A calculator can hide a wrong formula behind a polished interface. Have the relevant business team check every calculation.
- Publishing means exposing. A tool shared by link is often accessible to anyone who has the link. Do not put customer or financial data in it without access control.
- A tool without an owner dies. Ten employees each building their own tools means ten tools that nobody maintains. Assign an owner to every tool you keep.
- A prototype is not production. As soon as a tool has to connect to your ERP or CRM, manage permissions or handle the load, you need a developer.
Which model should you choose to produce artifacts?
Cost is not a criterion: less than one cent per artifact. Choose based on two questions:
- Is the artifact simple and disposable? (a diagram for a meeting) → a fast model is enough.
- Does the artifact have to work on the first try? (a calculator your sales reps will use in front of a customer) → pick a model known for the quality of its code, and test it.
Being able to switch models depending on the artifact is a real advantage. A multi-model platform like GENIAL gives access to more than 30 models in a single interface, without stacking up subscriptions. The principle is explained in our article on AI routers and multi-model platforms.
Where should you start?
Pick a diagram or a small tool that a team has been missing for a long time, and have the AI produce it in half an hour. To train your teams to write this kind of request, the Genial Academy offers learning paths and prompt libraries. For your overall tool stack, see our selection of the best AI tools for SMBs, and to identify your priorities, take the free AI self-assessment. The general method is in choosing an AI at work with benchmarks.
Erwan Simon is CEO and co-founder of GENIAL, a Bordeaux-based company that specializes in the operational rollout of generative AI in SMBs and mid-market companies. He is an accredited AI Expert with Bpifrance (France's public investment bank) and an ambassador of "Osez l'IA," the French government's program to help businesses adopt AI.
Frequently asked questions
What is an AI-generated artifact?
An AI-generated artifact is a usable object produced by a text model, beyond a written answer: a diagram, a dashboard, a web page, a calculator or a small game. The model writes the code (SVG, HTML, JavaScript) that produces the object, which you can then display, share or edit.
Can an AI build an internal tool without a developer?
Yes, for a simple tool: a quote calculator, an interactive checklist, a simulator, a form. According to the 2026 OpenRouter benchmarks, some models build a playable game from a single prompt, for $0.003 with gpt-oss-120b. For a tool that handles sensitive data or connects to your IT systems, you still need a developer for security and maintenance.
How much does it cost to generate a diagram or a prototype with AI?
Almost nothing. In October 2026, OpenRouter measures a drawing at $0.000032 with Mistral Nemo and a playable game at $0.003 with gpt-oss-120b. The real cost is the time spent writing the request, checking the result and adjusting it.
What are the risks of tools built by AI?
The three main risks are undetected calculation errors, data leaks when a tool is published without access control, and a growing pile of tools that nobody maintains. Have a business expert check every formula, never publish a tool containing internal data without access control, and assign an owner to every tool you keep.
Take action on your AI strategy
The GENIAL AI self-assessment measures your AI maturity and gives you prioritized use cases in under 5 minutes. Free, no commitment.
Related articles
Reasoning AI: When to Pay for a Model That Thinks
Gemini 4 Argon scores 96.5% on scientific reasoning. Does that mean you need a reasoning model everywhere? The tasks that justify one, and the ones that don't.
Generative AI for Image, Video and Voice: Which to Choose?
Recraft, FLUX, Grok Imagine, ElevenLabs, Voxtral: which generative AI for visuals, videos and voices? Measured 2026 costs and speeds, and pitfalls to avoid.
AI Agents: Choosing the Right Model for Business Processes
An AI agent that acts in your tools must follow your rules. What the 2026 τ²-Bench benchmark reveals, and a method to pick the right model for your agents.