Nanbeige4.3-33B
33B dense open-weight reasoning and coding model with a 512K context window, released under Apache 2.0.
Overview
Nanbeige4.3-33B is a 33-billion-parameter text-generation model from the Nanbeige LLM Lab, published on Hugging Face under the Apache-2.0 license as a fine-tune of Nanbeige4.3-33B-Base. It supports up to 524,288 tokens of context and ships BF16 safetensors. The model card reports 80.9 on Terminal-Bench 2.1, 65.5 on DeepSWE, 60.3 on SWE-Bench Pro, 89.5 on GPQA Diamond and 95.2 on AIME 2026, and acknowledges that Qwen3.8-27B scores higher on SWE-Bench Pro and GDPval and Gemma4-31B scores higher on GPQA Diamond. These figures are vendor-reported and had not been independently reproduced at the time of writing.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- 33B parameters, BF16 safetensors, Apache-2.0 license
- 524,288-token (512K) context window
- Fine-tuned from Nanbeige4.3-33B-Base
- Vendor-reported 80.9 on Terminal-Bench 2.1 and 65.5 on DeepSWE
- Vendor-reported 95.2 on AIME 2026 and 89.5 on GPQA Diamond
- Serving examples on the model card use 4-way tensor parallelism
Use cases
- Self-hosted agentic coding and terminal automation
- Long-context repository and document analysis
- Math and science reasoning on owned infrastructure
Pricing
Free open weights under Apache-2.0; no first-party hosted pricing published. Self-hosting needs multi-GPU hardware (model card examples use 4-way tensor parallelism).
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- Not listed
- Record updated
- 26 Oct 2026
- Record last checked
- 26 Oct 2026