# Best AI model under 32B for reasoning — October 2026

As of Oct 15, 2026, Gemma 4 31B & Qwen3 8 27B & Gemma 4 26B A4b are tied for state of the art in Reasoning, models at or under 32B (since Oct 8, 2026; sources LMArena Text (style-controlled, hard prompts); methodology v1.0).

Category board: https://isitsota.com/c/reasoning?slice=le-32b

## Top 10

- #1 Gemma 4 31B
- #1 Qwen3 8 27B
- #1 Gemma 4 26B A4b
- #2 Qwen3 5 27B
- #3 Qwen3 30B A3b Instruct 2507

## Crown changes

- Oct 8, 2026 — Reasoning: Gemma 4 31B and Qwen3 8 27B and Gemma 4 26B A4b are co-SOTA

## Weights

- Humanity's Last Exam: primary, weight 1
- GPQA Diamond: primary, weight 1
- LMArena Text (style-controlled, hard prompts): primary, weight 1
- LiveBench: secondary, weight 0.35