Nanbeige4-3B

Released
2025-12-13
Last refreshed
2026-09-07
Status
Researched 162d ago
Open sourceCommercial use: permittedJSON / Tool useChina

Nanbeige4-3B is a released json / tool use model with open-source; evaluate it while provider pricing coverage matures.

Nanbeige LLM Lab releases · 3 in the last 12 monthsChangelog →
Specifications
Family
Nanbeige4
Released
2025-12-13
Context
64K
Parameters
3B
Architecture
Decoder Only
Openness
Open source
License
Apache 2.0OSI-approvedPermitted
Weights
Available
Code
Unknown
Training
Pretrained
Created by
Nanbeige LLM Lab

Massive multilingual model, open-source focus.

Beijing, China
Founded 2023
Website
Pricing

No tracked provider pricing is available yet.

Links
Website HuggingFace

About

Nanbeige4-3B is an open-source 3B-parameter language model by Nanbeige LLM Lab (BOSS Zhipin), released December 2025. Pre-trained on 23 trillion high-quality tokens with SFT on 30M+ diverse instructions. Context extended to 64K via Adjusted Base Frequency (ABF). Sets state-of-the-art on AIME 2024 (90.4), AIME 2025 (85.6), and GPQA-Diamond (82.2) for sub-10B models, outperforming models up to 10× larger including Qwen3-32B. arxiv: 2512.06266. HuggingFace: Nanbeige/Nanbeige4-3B-Base.

Providers

No active provider routes are tracked for this model yet.

Capabilities

Reasoning

Benchmark peer barsfor JSON / Tool use

BFCLRank 23 of 58
Claude Opus 4.5
77.5
Claude Sonnet 4.5
73.2
Gemini 3 Pro
72.5
GLM-4.6
72.4
Nanbeige4-3Bcurrent
51.4

Benchmark scores(1)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionEvaluationSource
BFCL
51.4v4Observed 2026-10-26
Configuration: Nanbeige4-3B-Thinking-2511 (FC)
Harness: BFCL V4 official evaluation (gorilla berkeley-function-call-leaderboard), FC mode
Evaluator: Gorilla LLM team (UC Berkeley)
Confidence: confirmed
Notes: BFCL V4 leaderboard (Last Updated 2026-04-12), entry 'Nanbeige4-3B-Thinking-2511 (FC)': rank 25, Overall Acc 51.40%; Multi Turn 51.12%, Web Search 21.50%, Memory 36.77%, Non-Live AST 81.58%, Live 79.42%, Irrelevance 83.09%. Retrieved 2026-10-26.
Source

Migration checks

No linked migration route is available for this model yet.

API versions

nanbeige4-3b-basenanbeige4-3b-thinking-2511