All insights

Weekly9 October 20261 min read

This week in AI: small models do most of the work

Anthropic released Claude Haiku 5.5 and OpenAI's GPT-6 reached all ChatGPT users. The flagships get the headlines, but small, fast models handle most everyday tasks.

What happened

On 7 October, Anthropic released Claude Haiku 5.5, the newest of its small, fast models. The same week, OpenAI's GPT-6 models reached all ChatGPT users, after a first release to paying customers in September.

Our take

Flagship models get the attention, but in the services we build, small models do most of the work. Sorting incoming messages, pulling dates and names out of a form, checking a letter against a template, summarising a short case note: these tasks need speed, low cost and consistency more than deep reasoning.

Small models have other advantages for public services. They respond quickly, which matters when a resident is waiting. They cost a fraction of a flagship, which matters when a service handles thousands of requests a day. And some can run on your own infrastructure, which helps when data has to stay in a particular place.

A good design uses both. A small model handles routine work and passes harder cases to a more capable model, or to a person. Getting that split right, and testing it, often saves more money than any negotiation over price.

Keep reading

Related insights