Model
Model explorer

StarCoder2 3B

OPEN
BigCode · StarCoder2 family · released Feb 28, 2024

ServiceNow-trained 3B StarCoder2 for laptop/edge code generation; outperforms original 15B StarCoder on many benchmarks.

ReasoningCodingVisionFunction callingTool useAgentic
153.0
Elo · rank #373
Parameters
3B
Active params
3B (dense)
Context
16K tokens
Architecture
Transformer decoder, grouped-query + sliding-window attention (4K window/16K ctx), FIM; ServiceNow-trained
License
BigCode OpenRAIL-M
Languages
17+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
DS-1000Coding25.0%#4
best: StarCoder2 15B · 33.8%
GSM8KMath27.7%#156
best: Llama 3.1 405B · 96.8%
HumanEval+Coding27.4%#36
best: Mistral Small 3.2 (24B) · 92.9%
HumanEvalCoding31.7%#150
best: Claude Opus 4.5 · 99.4%
MBPP+Coding47.4%#36
best: Llama 3.1 405B · 88.6%
MBPPCoding57.4%#61
best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%
RepoBenchCoding35.5%#7
best: DeepSeek Coder 33B · 41.9%
Run it locally
VRAM @ Q4
2.1 GB
VRAM @ FP16
6 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF
StarCoder2 family
Elo progression across releases
StarCoder2 15BFeb 2024416.5StarCoder2 3BFeb 2024153.0
Resources
HFbigcode/starcoder2-3bGHbigcode-project/starcoder2DOCSbigcode-project.org/docs/about/models
API price weights · each benchmark row carries its own source badge (see methodology)