AI in RhinoArtisan
Choosing Models
Every model sits somewhere on a triangle of quality, speed and cost, and you cannot have all three at once. The strongest models are slower and more expensive; the fast cheap ones give up accuracy on hard problems.
There is no universally correct answer. There is a correct answer per job.
The models in RhinoArtisan
Each tool that runs on your OpenRouter key has its own model setting:
| Setting | Where | Used by | Default |
|---|---|---|---|
| AI model | Options → Artisan → Assistant | The Assistant | Claude Opus 4.8 |
| Flow Studio model | Options → Artisan → Integrations | Flow Studio AI | Claude Opus 4.8 |
The Assistant page also has a model per specialised job — a router that reads each message first, vision for the pictures you attach and reasoning for multi-step builds — plus the image models it uses through fal.ai. Their defaults are sensible: the router uses a small, fast model, and vision and reasoning follow the AI model. See Assistant options.
The AI model and Flow Studio dropdowns offer the same curated list. The dropdown shows each model by its OpenRouter id:
| Provider | Model | Id in the dropdown |
|---|---|---|
| Anthropic | Claude Opus 4.8, the default | anthropic/claude-opus-4.8 |
| Anthropic | Claude Opus 5 | anthropic/claude-opus-5 |
| Anthropic | Claude Fable 5 | anthropic/claude-fable-5 |
| Anthropic | Claude Sonnet 5 | anthropic/claude-sonnet-5 |
| xAI | Grok 4.5 | x-ai/grok-4.5 |
| OpenAI | GPT-5.6 Sol | openai/gpt-5.6-sol |
| DeepSeek | DeepSeek V4 Pro | deepseek/deepseek-v4-pro |
| Qwen | Qwen3 Coder Plus | qwen/qwen3-coder-plus |
| Mistral | Codestral | mistralai/codestral-2508 |
Every model on the list works with both tools; the difference is speed, depth and what each provider charges your OpenRouter account.
Two tools choose their models elsewhere:
- File Search uses neither setting. Its ✦ Smart mode always runs on a small, fast model (Claude Haiku), because turning a description into search filters is a short job.
- Generative AI Studio picks the image, video or 3D model per generation, from the mode chip next to the prompt. See Modes & Models.
Two settings, on purpose
The Assistant and Flow Studio AI get a model each because they do different jobs. That is deliberate, and it is the single most useful lever you have.
The Assistant works on your design. It answers questions about the piece, and it changes and builds it in several steps: “resize it, change the metal and give me the new price” is three steps that all have to be right. It defaults to Claude Opus 4.8, a frontier model, because the requests jewelers bring are rarely as simple as they look. If you mostly ask short, factual questions, a cheaper model can be enough: DeepSeek V4 Pro is the value pick, close-to-frontier answers at a fraction of the cost.
Flow Studio AI writes code. Volume is low and the stakes are high: a script that is subtly wrong wastes far more of your time than the price difference between models. This is where a frontier model earns its cost. Set it to the strongest one you are willing to pay for.
When to move up
Reach for a stronger model when the job has any of these:
- A long instruction with several conditions. Weaker models drop constraints halfway through.
- Real consequences. Anything you will act on in production.
- Code. The gap between models is widest here.
When to move down
Drop to something cheaper and faster when:
- You are asking short, factual questions in a long session.
- You are exploring, and a rough answer is enough to point you in a direction.
- You are repeating a task you have already validated.
Generative AI Studio works the same way
The image, video and 3D models in Generative AI Studio sit on the same triangle. Use a fast, inexpensive model while you are exploring composition and framing, then regenerate the one you like on a stronger model for the final image. Paying the top rate for a draft you will discard is the most common way to overspend.