Skip to content

LLM Model Selection

Pick which OpenAI model powers each task in your account. The defaults are sensible — change them only when you have a specific reason (cost, latency, or quality on a particular task type).

Available Models

All models are OpenAI-hosted. Functional AI proxies the call so you don't need an OpenAI key.

Model Best For Context Notes
gpt-5-mini (default) Customer-facing chat, function calling Standard Economical small GPT-5 model — fast and cost-efficient
gpt-4.1-mini High-volume tasks, summaries 1M tokens Balanced model for speed and intelligence
gpt-4.1-nano Cheapest, simple/batch work Standard Fastest inference, lowest cost
gpt-4o Hardest reasoning, technical edge cases 128K tokens High-intelligence model, premium price
gpt-4o-mini Low-cost everyday tasks 128K tokens Affordable small model

OpenAI only, today

Functional AI uses OpenAI as the underlying LLM provider. We don't currently route to Anthropic or Google models in production. If you need a different provider, contact support.

Per-Task Configuration

Each task can use a different model. Settings are account-wide (they apply to every assistant you own).

Task What It Does Sensible Default
Chat Logged-in dashboard chat gpt-5-mini
Shared Chat Public widget on Shopify, websites, share links gpt-5-mini
Summary Conversation summaries and titles in the inbox gpt-4.1-mini
QA Generation Generates Q&A pairs from knowledge-base files for RAG gpt-4.1-mini
Function Creation Generates Python for custom functions gpt-5-mini
Function Test & Fix Iterates to debug failing custom functions gpt-4.1-mini

New accounts start with gpt-5-mini for every task. The recommendations above are a cost-tuned starting point — keep customer-facing tasks (Chat, Shared Chat, Function Creation) on a stronger model and downgrade the offline/batch tasks.

Configuring

  1. Go to Account → LLM Settings
  2. For each task, pick the provider (OpenAI) and model
  3. Click Save Changes — changes apply to new messages immediately. Conversations already in flight finish on their previous model.

Test before rolling out

After changing the Shared Chat model, run a few realistic visitor questions through the widget before announcing anything. Lower-cost models can miss nuance on product-specific phrasing.

Cost vs. Quality Trade-offs

Stay on gpt-5-mini when:

  • The task is customer-facing (Chat, Shared Chat)
  • Function calling needs to be reliable
  • You need consistent JSON / structured output

Downgrade to gpt-4.1-mini or gpt-4.1-nano when:

  • The task is internal (Summary, QA Generation, Function Test & Fix)
  • The work is high-volume and quality differences are small
  • You're optimizing for cost during development

Upgrade to gpt-4o when:

  • A specific assistant repeatedly fails on complex reasoning even with good knowledge-base coverage
  • You're willing to pay materially more per message for better answers

You can compare token pricing on the OpenAI pricing page. Functional AI bills you per message at your tier rate (see Billing) — the model only changes our underlying cost, not your subscription cost. Overflow is a flat $0.10/message regardless of model.

Common Questions

Can I use different models for different assistants?

No. Model selection is account-wide. All your assistants share the same Chat / Shared Chat / etc. models.

Does the model affect my message quota?

No. Each AI response counts as 1 message regardless of which model produced it.

What happens if a model is deprecated?

You'll get advance notice and your settings will auto-fall-back to a comparable model (most likely the current default).