Subscribe

Nanbeige4.3-33B

Nanbeige released Nanbeige4.3-33B on Hugging Face under Apache 2.0, a 33B model with a 512K token context window aimed at agentic coding.

It is the large model of a planned family, which will later be distilled into smaller models as Nanbeige did with its 3B line earlier this year. Per its own model card it scores 65.5 on DeepSWE against 42.5 for Qwen3.8-27B. It trails Qwen narrowly on SWE-Bench Pro and trails Gemma4-31B on GPQA Diamond.

Date
Sunday 25 October 2026
Lab
Nanbeige
Kind
open-weights
Access
open weights

Figures

MeasureValueMeasured by
DeepSWE65.5
Qwen3.8-27B scored 42.5
company
Automation Bench54.2
Qwen3.8-27B scored 38.3
company
JobBench44.8
Qwen3.8-27B scored 34.5
company
AIME 202695.2
Qwen3.8-27B scored 94.5
company
GPQA Diamond89.5
Gemma4-31B scored 91.3
company
SWE-Bench Pro60.3
Qwen3.8-27B scored 61.7
company

Release date is given only as this week in the coverage, so 2026-10-25 is the article date. Benchmarks come from the lab's model card as reported by Startup Fortune.

Sources

  1. startupfortune.com/nanbeiges-new-33b-open-model-takes-aim-at-qwen-and-gemma-on-coding/

This record was partly confirmed: some claims could not be checked on 25 October 2026. How we check

Read the daily brief for 25 October 2026