AI Models
Hub iAbot isn't a chatbot with a single fixed model. It's a multi-model orchestrator that automatically selects the most suitable artificial intelligence for each task: writing React code, designing UI, generating images, transcribing voice, or planning architectures.
Hub AI models
price, quality, and best use
Best overall pick
GPT-5.6 Luna Pro
$0.20 / $1.20 regular · daily use
Best full-stack
GLM 5.2
Complete apps · logic and long context
Hard bugs
Grok 4.6
Architecture, databases, and persistence
Maximum quality
GPT-5.6 Sol Pro
$5 / $30 · last resort only
DeepSeek V4 Flash 0731
GoodInput / 1M: $0.08 · Output / 1M: $0.252
Ask, analysis, and small changes.
GPT-5.6 Luna Pro
Best valueInput / 1M: $0.10 promo ($0.20 regular) · Output / 1M: $0.60 promo ($1.20 regular)
Default model for almost everything
DeepSeek V4 Pro 0813
HighInput / 1M: $0.435 · Output / 1M: $0.87
Affordable reasoning and backend.
GLM 5.2
Very highInput / 1M: ≈ $0.50 · Output / 1M: ≈ $3.15
Full apps, logic, and full-stack.
Gemini 3.7 Flash
HighInput / 1M: $0.75 · Output / 1M: $3.75
Multimodal, images, and mobile.
Muse Spark 1.2
Very highInput / 1M: $1.25 · Output / 1M: $4.25
Design, frontend, and polished UI.
Grok 4.6
ExcellentInput / 1M: $2.00 · Output / 1M: $6.00
Hard bugs, architecture, and databases.
Qwen 3.8 Max
ExcellentInput / 1M: $2.00 · Output / 1M: $6.00
Agents and autonomous reasoning.
Claude Sonnet 5
ExcellentInput / 1M: $2.00 · Output / 1M: $10.00
Delicate repairs and long projects.
GPT-5.6 Sol Pro
MaximumInput / 1M: $5.00 · Output / 1M: $30.00
Last resort: maximum quality, maximum cost
Luna currently has a 50% discount on OpenRouter. Pro mode uses more reasoning tokens.
USD · base prices · Hub may add orchestration costs
How Hub decides which model to use
You describe the outcome; Hub picks the intelligence.
You don't need to be a prompting expert or know which model is best for each task. Hub's engine analyzes your request and routes the task to the optimal model (or a combination of several).
| Layer | What it does | Typical models |
|---|---|---|
| AppWeb Studio | Generates React, SQL, fixes errors, and improves UI | DeepSeek V4, Qwen3, Kimi |
| Builder Agent (Flow) | Generates ATZ flows from text | DeepSeek V4 + Groq |
| Runtime / Resources | What your app uses in production | OpenRouter (FLUX.2, Seedance, DeepSeek, Kokoro…) |
| Channels | Automatic STT on WhatsApp/Telegram | Groq Whisper |
AppWeb Generation (Studio Canvas)
In Hub Studio, websites are created only with Canvas + Copilot. Depending on the chat mode (Ask, Plan, or Agent) and the task, Hub routes work to different models. In the composer you can also pick the model for the turn (default GPT-5.6 Luna Pro; also DeepSeek V4 Flash/Pro, GLM 5.2, Grok 4.6, Sol Pro, and the rest of the guide) and Edit reasoning (Off / High / XHigh depending on model). Usage details in the workspace.
Typical routing layers:
1. Ask and Plan
High-reasoning models investigate the workspace, clarify requirements, and propose structured plans before editing code.
2. Agent (build)
Code-optimized models apply edits on the live workspace: files, React components, and dependencies.
3. Vision and attachments
When you attach screenshots or PDFs, vision-capable models interpret the reference to align design and content.
4. Repair and iteration
If a build fails or you ask for a deep change, the agent retries with reasoning models until the preview is stable.
Learn more about the AppWeb ecosystem
Automations and Hub Flow
When building automated conversational flows, the Builder Agent takes your text instruction and turns it into functional nodes (triggers, conditionals, AI actions). To do this quickly and accurately, Hub uses a "multi-model combo":
- Complex nodes and logic: Resolved by robust models like DeepSeek V4 Pro or Flash.
- Simple nodes and fast structure: Delegated to extremely fast models served on Groq.
Multimodal Resources in Production
Once your app or bot is live and interacting with real users, the resources it integrates use calls to specialized APIs:
Via OpenRouter: FLUX.2 Pro/Klein, Seedream 4.5, Seedance 2.0 and Veo 3.1 Lite.
Chat and analysis via OpenRouter: DeepSeek V4 Pro and Flash, Kimi K3, Claude Sonnet 5, MiMo V2.5.
Whisper Large V3 Turbo (STT) and TTS with Kokoro 82M or Gemini 3.1 Flash, all via OpenRouter.
Invoice reading (OCR) and PDF search with Google technologies and RAG pipelines.
View the full resource catalog
Implicit Voice: Speaking is natural
On conversational channels like WhatsApp and Telegram integrated via Hub Flow, you don't need to configure voice recognition nodes.
If a user sends a voice note, Hub intercepts it and uses Groq Whisper (whisper-large-v3-turbo) to transcribe it into text almost instantly before it reaches your flow. Your bot always "reads" text, greatly simplifying the logic of your automations.
This same concept applies in Hub's web panel, where the built-in microphone lets you dictate natural instructions natively.
Models and Hub AI Credit
One euro balance — Hub AI Credit — covers Copilot in Studio and AI in your published apps:
Ask, Plan, Agent, code generation, images (FLUX.2), video (Seedance), or LLM chat in a published app debit this balance based on the model and request size. If the balance reaches zero, the app stays online; only AI actions pause.
Every plan includes Hub AI Credit (e.g. Trial 4 €). You can top up 1:1 with a card. See the Hub AI Credit guide for details.
Guide: Hub AI Credit • View plans and pricing
Providers and Partners
We route requests to the best available infrastructure, working with industry leaders to ensure speed and reliability: