Model
Model explorer

BLOOMZ (176B)

OPEN
BigScience · BLOOMZ family · released Nov 3, 2022

BLOOM 176B fine-tuned on the xP3 crosslingual task mixture for zero-shot multilingual instruction following.

ReasoningCodingVisionFunction callingTool useAgentic
-106.4
Elo · rank #413
Parameters
176.25B
Active params
176.25B (dense)
Context
2K tokens
Architecture
Dense decoder-only transformer (BLOOM architecture), instruction fine-tuned on xP3
License
BigScience RAIL License v1.0
Languages
46+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ANLIReasoning48.6%#4
best: Phi-3-small (7B) · 58.1%
ARC-ChallengeReasoning55.4%#108
best: Llama 3.1 405B · 96.9%
HellaSwagReasoning75.2%#112
best: Claude 3 Opus · 95.4%
HumanEvalCoding12.1%#181
best: Claude Opus 4.5 · 99.4%
MMLUKnowledge47.7%#239
best: OpenAI o3 · 92.9%
SuperGLUEReasoning82.1%#3
best: UL2 20B · 90.7%
TruthfulQAKnowledge43.9%#68
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
WinoGrandeReasoning59.3%#133
best: PaLM 2 · 90.9%
XWinogradReasoning69.1%#7
best: Sarashina2-70B · 91.8%
Run it locally
VRAM @ Q4
106 GB
VRAM @ FP16
352 GB
Fits on (Q4)
M3 Ultra 512GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
INT8 (bitsandbytes)
BLOOMZ family
Elo progression across releases
BLOOMZ (176B)Nov 2022-106.4
Resources
HFbigscience/bloomzAPIAvailable via BigScience API
API price weights · each benchmark row carries its own source badge (see methodology)