Leaderboard / Nanbeige (Kanzhun)
Nanbeige4.3 33B
Nanbeige (Kanzhun) · Nanbeige4 · released 2026-10-24 · data checked 2026-11-01
Nanbeige4.3 33B is a 33.1B-parameter dense model from Nanbeige (Kanzhun) (China), released 2026-10-24 under the Apache-2.0 license. It scores 77.8 on the Small LM Index, #1 of 34 medium models. It needs about 23 GB of memory at 4-bit with an 8K-token context.
Vendor-reported scores, unusually high for its size; treat with care until independently reproduced.
Benchmarks
Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 4 reported by Nanbeige (Kanzhun). How we label sources.
Not reported: estimated
What the index model expects from this model's other scores. Not a measurement.
Specifications
- Lab
- Nanbeige (Kanzhun) (CN · China)
- Architecture
- Dense transformer
- Total parameters
- 33.1B
- Active parameters
- All (dense)
- Size class
- Medium
- Context window
- 524K tokens
- Input modalities
- text
- Reasoning
- Reasoning (thinking)
- License
- Apache-2.0
- Openness
- Open weights
- Memory to run
- ~23 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 17.8K downloads (30 days) · 1K likes
Compare with
- Nanbeige4.3 33B vs Qwen3 32B (32.8B, index 52.4)
- Nanbeige4.3 33B vs DeepSeek-R1-Distill-Qwen-32B (32.8B, index 50.9)
- Nanbeige4.3 33B vs Sarvam 30B (32.2B, index 61.6)
- Nanbeige4.3 33B vs OLMo 3.1 32B Think (32.2B, index 57.0)
Ranked in
- #3 in Best small open models for math
- #1 in Best LLMs for a 24 GB GPU (RTX 4090, RTX 3090)
- #1 in Best open models with an Apache 2.0 or MIT license
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vllm serve Nanbeige/Nanbeige4.3-33BMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 256K context |
|---|---|---|---|
| 16-bit F16 | 68 GB | 74 GB | 123 GB |
| 8-bit Q8_0 | 37 GB | 43 GB | 92 GB |
| 4-bit Q4_K_M | 22 GB | 28 GB | 77 GB |