LLM Reference

Nanbeige4 Models by Nanbeige LLM Lab

2 models2025–2026Up to 256k ctx

Details

Models2
Released2025–2026
Max context256k

Capabilities

ReasoningAll models
Function Calling1 of 2 models
Tool Use1 of 2 models

About

Nanbeige4 is the fourth-generation model series from Nanbeige LLM Lab (BOSS Zhipin / Kanzhun Limited). Characterized by strong reasoning at small scale: 3B parameters pre-trained on 23 trillion tokens (Nanbeige4-3B), and the enhanced Nanbeige4.1-3B with 256K context and agentic capabilities. Both models are open-source and available on HuggingFace.

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

2 in view

Use when the workload needs 256k context, 4B parameters, and reasoning.

2026-02256k context4B parametersreasoning

Use when the workload needs 64k context, 3B parameters, and reasoning.

2025-1264k context3B parametersreasoning

Release Timeline

2 release groups
2026-02
1 current
Nanbeige4.1-3B
256k context4B parametersreasoning
Current
2025-12
1 current
Nanbeige4-3B
64k context3B parametersreasoning
Current

Specifications(2 models)

Nanbeige4 model specifications comparison
ModelReleasedContextParametersReasoningFn CallingTool Use
Nanbeige4.1-3B2026-02256k4BYesYesYes
Nanbeige4-3B2025-1264k3BYesNoNo

Frequently Asked Questions

What is Nanbeige4 used for?
Nanbeige4 is used for china, reasoning, and agent workflows and tool use. The family description and listed model capabilities point to those workloads as the best fit.
How does Nanbeige4 compare to SenseNova U1?
Nanbeige4 by Nanbeige LLM Lab is strongest where you need china, while SenseNova U1 by SenseTime is the closest related family to check for china. Nanbeige4 has 2 listed variants and reaches up to 256k context, so compare the specs and pricing tables before choosing a production model.
Which Nanbeige4 model should I use?
If price is the main constraint, use the pricing table first because Nanbeige4 does not have complete provider pricing in the local data. For the most capable/latest local choice, evaluate Nanbeige4.1-3B with 256k context and reasoning, tool use, and function calling.