Model upgrades and smarter file handling

Faster and more efficient Handlers, and more reliable text extraction from large and complex PDFs.

On this page

Release notes for 20 October–2 November 2026

Our biggest set of model upgrades in a long time

This is our biggest set of model upgrades in a long time, and it’s rolled out to all customers. Your Handler is faster and more efficient — and there’s a bigger story behind these upgrades than new models alone.

For much of 2026, Anthropic’s Claude models led the way, striking a balance of intelligence and cost that was right for the time. More recently, a wave of highly capable models has hit the market — both open-weight and proprietary.

For the first time at Handle, we’ve moved a portion of Handler workloads onto open models. We’re excited about this step: we see open models as the future of Handle. We’ve also partnered with a new API provider to improve our access to open-weight models while maintaining our standards of speed and reliability.

OpenAI is competitive again on Handle’s benchmarks, too — and we’re routing a portion of the workload there once again.

For you, these upgrades mean less time waiting for work to come back. More efficient Handlers get more out of the capacity they have — you should expect your Handler’s capacity to go further.

Smarter file handling

When you ask your Handler to work with a PDF — summarising a long report, pulling the numbers out of an invoice — it starts by reading the text out of the document. Everything your Handler does with the contents draws on that first step.

Handle has been able to read the text out of your PDFs for a while now. We’ve added tooling to make that extraction more consistently accurate — sharper and more reliable, including for large and complex documents like multi-page reports and invoices.

The more faithfully the text comes out, the sounder the starting point for everything that follows — the summaries, the answers, the follow-up work.

Alongside this, a reminder from Introducing artifacts: Handlers can examine older Excel (.xls) spreadsheets and extract their contents.


Questions or feedback? We’d like to hear from you.