---
title: "Open-weight models close the gap on reasoning benchmarks"
description: "Three releases in three weeks, all small enough to run on a single GPU. Here is what changed and what it costs to serve."
canonical_url: "https://flexitech.io/en/article/open-weight-models-close-the-gap"
publisher: "Flexitech Digest"
author: "Flexitech Digest"
section: "AI"
date_published: "2026-09-04T02:29:24.074Z"
date_modified: "2026-09-09T02:29:25.953Z"
language: en
read_time_minutes: 4
---

# Open-weight models close the gap on reasoning benchmarks

Three releases in three weeks, all small enough to run on a single GPU. Here is what changed and what it costs to serve.

_By Flexitech Digest · 4 Sep 2026_

## What to know

- Three open-weight models now fit on a single accelerator.
- The reasoning-benchmark gap with frontier models has narrowed enough to re-run your own evaluations.
- Serving cost, not capability, is now the deciding factor for most workloads.

Three open-weight releases landed inside a month, and each one fits on a single accelerator. The benchmark gap that justified paying for a frontier model has narrowed to the point where it is worth re-running your own evaluations before renewing.

---

Published by Flexitech Digest. Canonical version: https://flexitech.io/en/article/open-weight-models-close-the-gap