Model guide

Recommended AI models: which one to use for which job.

A quick-reference guide to the AI models we recommend for different kinds of work, so your staff always know which one to reach for.

Talk to us AI for banking
The short version
  • Start here: ChatGPT Terra or Claude Sonnet for everyday work
  • Hardest problems: ChatGPT Sol or Claude Opus
  • Quick, high-volume tasks: ChatGPT Luna or Claude Haiku
  • Background tasks: GPT Mini
OpenAI Anthropic

Deep thinking

Tackle the hardest problems

OpenAI

ChatGPT Sol

OpenAI

For the hardest problems, with detailed analysis and multi-step reasoning. Trades speed for depth.

Reasoning
Speed
Knowledge cutoffFeb 2026
Anthropic

Claude Opus

Anthropic

Deep reasoning with excellent writing. Strong for complex drafting and careful review.

Reasoning
Speed
Knowledge cutoffMay 2026

Examples at your bank or credit union: reviewing a long policy for gaps, or drafting a careful audit response.

Everyday work

Get day-to-day work done

Start here
OpenAI

ChatGPT Terra

OpenAI

The everyday default, for strong reasoning at a quick pace. Handles most day-to-day work.

Reasoning
Speed
Knowledge cutoffFeb 2026
Anthropic

Claude Sonnet

Anthropic

A balanced workhorse with excellent writing. A good choice if you prefer Anthropic's style.

Reasoning
Speed
Knowledge cutoffJan 2026

Examples at your bank or credit union: drafting member correspondence, summarizing meeting notes, or answering questions in customer support.

Fast & light

Move fast on simple tasks

OpenAI

ChatGPT Luna

OpenAI

Built for speed. Great for quick questions, summaries, and high-volume tasks.

Reasoning
Speed
Knowledge cutoffFeb 2026
Anthropic

Claude Haiku

Anthropic

Very fast and capable. Ideal for drafting, quick lookups, and routine tasks.

Reasoning
Speed
Knowledge cutoffFeb 2025

Examples at your bank or credit union: quick lookups, short summaries, or high-volume work such as triaging customer complaints.

General use provider

For system operations like thread summarization and prompt improvement, we recommend GPT Mini. It's fast and efficient for these background tasks.

Knowledge cutoffs are approximate and mark where a model's trained knowledge ends. Models with web search enabled can still answer about more recent events.

In the portal

Switch models in one click, or ask several at once.

Everything on this page is available from the same model menu. Your staff can move a conversation to a different model, or send one prompt to several and compare the answers.

Model picker

Every recommended model, one menu.

Internal models, your RAG assistants, and External models such as ChatGPT, Claude, and Gemini are listed together, grouped so staff can see which run inside your walls.

  • Grouped into Internal and External models
  • Only the models each group is allowed to use
  • New models appear here as soon as your IT team adds them
Side-by-side answers

See how each model handles the same prompt.

Run one prompt on ChatGPT, Claude, and Gemini and flip between their responses. It is the quickest way to find out which model suits a given task at your bank or credit union.

  • One prompt, several models
  • Responses shown as tabs to compare
  • The same DLP and logging on every model
Meeting summary
FileEditPromptAssistant
Multi Internal Use prompt ⌘ ↵

Summarize my notes from yesterday’s meeting with Bayside Logistics and generate a list of follow-up items with owners and due dates.

bayside-meeting-notes.docx AI can make mistakes. Verify important information with alternative sources.
Thought for a few seconds

Meeting summary

Bayside Logistics is looking to refinance a $4.2M equipment loan and open a $750K revolving line of credit to cover seasonal cash flow. Their CFO flagged turnaround time and online treasury access as the deciding factors, and they’re comparing us with one other lender.

Follow-up items

  1. Send the term sheet for the refinance and credit line — Jordan Ellis, Fri Oct 3
  2. Request three years of financial statements and the equipment list — Bayside CFO, Wed Oct 8
  3. Book a treasury management demo with their AP team — Priya Shah, next week
  4. Confirm underwriting timeline with credit — Jordan Ellis, Thu Oct 2

Want me to turn this into a follow-up email to the client?

Governed, whichever model you choose

Every model, behind the same controls.

In MultiModel, the model is a choice for your users. The guardrails are a choice for you, and they apply to every model in the portal.

Security & compliance

Access by group

Decide which groups get which models.

AI guardrails

Data loss prevention

Redact sensitive customer and member data on the device, before a prompt reaches any model.

Data loss prevention

One vendor, many models

Compare models on real performance and avoid lock-in to a single AI provider.

Vendor management
FAQ

Choosing an AI model: FAQ

Straight answers for CIOs, CISOs, and compliance officers evaluating AI.

Talk to us
Which model should your staff start with?

Start with an everyday model: ChatGPT Terra, or Claude Sonnet if you prefer Anthropic's writing style. They offer strong reasoning at a quick pace and handle most day-to-day work. Move up to a deep-thinking model for the hardest problems, or down to a fast model for quick, high-volume tasks.

Why recommend models from more than one provider?

Different models suit different work, and relying on a single AI provider creates concentration risk. MultiModel gives your staff several providers' models in one portal. See AI vendor management.

What is a knowledge cutoff?

It is the approximate point where a model's trained knowledge ends. A model will not know about events after its cutoff from training alone, but models with web search enabled can still answer about more recent events.

Which model runs background tasks like thread summaries?

For system operations such as thread summarization and prompt improvement, we recommend GPT Mini. It is fast and efficient for these background tasks.

Can we control which staff get which models?

Yes. In MultiModel you decide which groups get which models. See AI guardrails.

Is it safe to use these models with member or customer data?

Through MultiModel, every model runs behind the same controls. Sensitive data can be redacted on the employee's device before a prompt is sent, and prompts are excluded from model-provider training. See data loss prevention and AI for banking.

Give your staff the right model for every job.

Every model in this guide runs through the same guardrails in MultiModel. We'll show you how to offer them to the right teams.

Talk to us