Back to News

AI · Oct 10, 2026

Big model, small model: how to choose the right AI model for each task

On September 22, two of the biggest AI labs shipped on the same day. OpenAI introduced GPT-6 Sol and GPT-6 Luna, two models meant to bring frontier intelligence to everyday work at different balances of capability and cost. The same day, Anthropic launched Claude Opus 5.5, which it says performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5.

Both told the market the same thing: you do not need the most powerful model for most work.

The family-of-models pattern

Every major lab now sells a family of models rather than a single one. There is usually a top model for hard reasoning, a middle model for everyday work, and a small, fast model for high-volume tasks. Prices between the top and bottom can differ many times over.

OpenAI followed up a week later with GPT-6.1 Sol, which it says delivers near-Astra intelligence at one-fifth of Astra's standard API token prices. Anthropic followed with Sonnet 5.5 on September 28 and Haiku 5.5 on October 7.

For a business building AI into its operations, this is good news. It means you can pay for intelligence where it matters and save where it does not.

How to match a model to a task

Here is the simple framework we use when designing an automation.

Use a small, fast model when:

  • The task is repetitive and well defined, like sorting incoming email into categories
  • You are pulling specific fields out of a document, like an invoice number and total
  • Speed matters more than nuance, like a live chat that needs to answer in under a second

Use a middle model when:

  • The task needs some judgment, like drafting a customer reply that matches your tone
  • The work involves several steps but follows a familiar pattern
  • You need good writing quality at a reasonable cost

Use the top model when:

  • A mistake is expensive, like reviewing contract terms
  • The task is open-ended and requires planning, like researching a new market
  • Earlier attempts with smaller models failed

Mixing models in one workflow

The best automations often use more than one model. A small model reads every incoming support email and decides which ones are urgent. A middle model drafts replies to routine questions. Only the hard cases go to the top model, or to a person.

This is how you keep costs predictable as volume grows. If every email goes through the most expensive model, your bill scales with your inbox. If only the hard ones do, it scales with your complexity.

Watch out for lock-in

The other lesson from this month is how fast things change. Models that were top of the line in August were already mid-tier by October. If your automation is hard-wired to one specific model, every upgrade becomes a project.

Build so the model is a setting, not a foundation. That way, when a cheaper model gets good enough for one of your steps, switching is a configuration change and a test run, not a rebuild.

What this means for your business

  • Do not default to the most powerful model. Most business tasks do not need it.
  • Break workflows into steps and assign each step the cheapest model that does it well.
  • Test model changes on real examples before switching.
  • Keep your automations flexible so you can change models as prices drop.

Choosing models is one of the decisions that quietly determines whether an AI project stays affordable. If you want a second opinion on how your workflows are set up, we are happy to take a look.

Sources