Nanbeige4.3-33B
An open-source large language model released by the Nanbeige team with 33 billion parameters. It outperforms models of the same tier on multiple benchmarks and serves as the teacher-model foundation for a small-model distillation pipeline.
Timeline (last 90 days)
In the model card, the official recommendation is to use the preserve_thinking=true setting in multi-turn tool-calling scenarios.
On the SWE-Bench Pro benchmark, Nanbeige4.3-33B scores 60.3, slightly below Qwen3.8-27B's 61.7.
Nanbeige4.3-33B supports a 512K context window.
On the Terminal-Bench 2.1 benchmark, Nanbeige4.3-33B scores 80.9 and Qwen3.8-27B scores 73.0, both obtained under the same test framework and inference settings.
The model weights of Nanbeige4.3-33B are now available for download on Hugging Face.
Nanbeige4.3-33B is positioned as a teacher model to guide smaller agent models, including a planned 3B version.
Nanbeige4.3-33B reports a score of 80.9 on Terminal-Bench 2.1.
The Nanbeige LLM Lab released Nanbeige4.3-33B, a 33B dense model.
Nanbeige4.3-33B uses the Apache 2.0 license.
Nanbeige4.3-33B has a context length of 512K.
7 more timeline events
All Facts (17)
The model weights of Nanbeige4.3-33B are now available for download on Hugging Face.
50%UnverifiedIn the model card, the official recommendation is to use the preserve_thinking=true setting in multi-turn tool-calling scenarios.
50%UnverifiedNanbeige4.3-33B is positioned as a teacher model to guide smaller agent models, including a planned 3B version.
50%UnverifiedOn the SWE-Bench Pro benchmark, Nanbeige4.3-33B scores 60.3, slightly below Qwen3.8-27B's 61.7.
50%UnverifiedOn the Terminal-Bench 2.1 benchmark, Nanbeige4.3-33B scores 80.9 and Qwen3.8-27B scores 73.0, both obtained under the same test framework and inference settings.
50%UnverifiedNanbeige4.3-33B supports a 512K context window.
50%UnverifiedNanbeige4.3-33B reports a score of 80.9 on Terminal-Bench 2.1.
50%UnverifiedNanbeige4.3-33B uses the Apache 2.0 license.
50%UnverifiedNanbeige4.3-33B has a context length of 512K.
50%UnverifiedThe 512K context is the most direct product selling point of Nanbeige4.3-33B.
50%UnverifiedThe Terminal-Bench score of 80.9 currently comes from the official self-report, and third-party reproduction results have not yet appeared.
50%UnverifiedOfficial information on the specific distillation methods, data composition and timeline for Nanbeige4.3-33B remains limited.
50%UnverifiedThe specific benchmark items, evaluation configurations and comparison conditions for Nanbeige4.3-33B have not yet been fully presented in public information.
50%UnverifiedThe official model card explicitly states that Nanbeige4.3-33B will serve as the foundation for a subsequent small-model distillation pipeline, benefiting Nanbeige4.3-3B first.
50%UnverifiedAccording to the Reddit community citing the Hugging Face model card, Nanbeige4.3-33B outperforms Qwen3.8-27B on a series of benchmarks.
50%UnverifiedThe Nanbeige LLM Lab released Nanbeige4.3-33B, a 33B dense model.
50%UnverifiedThe Nanbeige4.3-33B model has been released on Hugging Face with 33 billion parameters.
50%