Model
Model explorer

Stable Beluga 2 (FreeWilly 2) 70B

OPEN
Stability AI (CarperAI) · Stable Beluga (FreeWilly) family · released Jul 21, 2023

Briefly #1 on the HF Open LLM Leaderboard (Jul 27 2023); Llama 2 70B fine-tuned on ~600K Orca-style synthetic examples.

ReasoningCodingVisionFunction callingTool useAgentic
511.0
Elo · rank #301
Parameters
70B
Active params
70B (dense)
Context
4K tokens
Architecture
Llama 2 70B dense Transformer, Orca-style supervised fine-tune
License
Stable Beluga Non-Commercial Community License Agreement
Languages
—
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
ARC-ChallengeReasoning71.1%#46
best: Llama 3.1 405B · 96.9%
DROPReasoning50.3#55
best: Hunyuan-T1 · 93.1
GSM8KMath35.9%#146
best: Llama 3.1 405B · 96.8%
HellaSwagReasoning86.4%#28
best: Claude 3 Opus · 95.4%
MMLUKnowledge68.8%#155
best: OpenAI o3 · 92.9%
TruthfulQAKnowledge59.4%#25
best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%
WinoGrandeReasoning83.0%#20
best: PaLM 2 · 90.9%
Run it locally
VRAM @ Q4
40 GB
VRAM @ FP16
140 GB
Fits on (Q4)
M3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGML · GGUF · GPTQ
Stable Beluga (FreeWilly) family
Elo progression across releases
Stable Beluga 1 (FreeWilly 1) 65BJul 2023177.8Stable Beluga 2 (FreeWilly 2) 70BJul 2023511.0
Resources
HFstabilityai/StableBeluga2
API price weights · each benchmark row carries its own source badge (see methodology)