How Exowatt outperformed frontier models on specific tasks with SageMaker serverless customization
Exowatt beats frontier models on accuracy with SageMaker serverless model customization
- Who
- Exowatt
- What
- shared a repeatable SageMaker AI serverless model customization pipeline for fine-tuning and evaluating task-specific models for edge and cloud deployment
- Quote
In this post, we show how Exowatt used Amazon SageMaker AI serverless model customization to build a repeatable AI pipeline.
What happened
Exowatt used Amazon SageMaker AI serverless model customization to fine-tune Qwen3-1.7B for edge alarm triage and Qwen3-14B for cloud energy-market analysis. The models achieved 0.911 and 0.975 macro-F1 on their respective tests, outperforming the reported frontier baselines. Its reusable pipeline combines curated data, checked paraphrases, fine-tuning, and evaluation, supporting offline CPU deployment at the edge and cloud inference at $0.60 per 1,000 answers versus $29.26 for Claude Opus 4.7.
Why it matters
The case illustrates a transferable fine-tuning workflow in which verified paraphrase diversity improves task accuracy while a shared training pipeline supports both offline edge inference and managed cloud deployment.
Read next
See the full storyExowatt Fine-Tunes Qwen3 Models for Alarm Triage and Energy-Market AnalysisThe story so far
Exowatt used Amazon SageMaker AI serverless model customization to fine-tune two Qwen3 models for alarm triage and energy-market analysis, with both outperforming the frontier-model baselines reported in an AWS Machine Learning Blog post. Qwen3-1.7B achieved 0.911 macro-F1 on the edge alarm-triage test, while Qwen3-14B achieved 0.975 macro-F1 on the cloud energy-market analysis test. Those results apply to the specific tasks tested. Exowatt’s reusable pipeline combines curated data, checked paraphrases, fine-tuning, and evaluation. It supports offline CPU deployment for the edge model and cloud inference for energy-market analysis. The post reported cloud inference costs of $0.60 per 1,000 answers, compared with $29.26 for Claude Opus 4.7.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-10-27