This 15B-parameter model matches the performance of giants like DeepSeek-R1-0528, Mistral-medium-1.2, and Gemini Flash 2.5 on the AAI 52 Index—but is 8-10x smaller!
-
⚡ Frontier-level reasoning on a single GPU.
-
📦 Open Weights for immediate evaluation.
-
🧠 Excellent reasoning scores (AIME2025: 88, GPQA: 71).
-
✅ Highly reliable instruction following.
-
🚀Nemotron Powered: ≈25% of its depth up-scaling data came from the NVIDIA Nemotron collection.
Try It
-
Chat Demo: Apriel Chat - a Hugging Face Space by ServiceNow-AI
-
Model Weights: https://nvda.ws/4pSKo1i
Tech Report: Apriel-1.5-Thinker.pdf · ServiceNow-AI/Apriel-1.5-15b-Thinker at main