AWS Brings OpenTSLM Time-Series Reasoning to SageMaker JumpStart

What happened
Amazon Web Services announced on November 17, 2026, that the OpenTSLM-Flamingo Gemma 3 4B time-series reasoning model is now available through SageMaker JumpStart. This model can be deployed as a real-time inference endpoint within AWS environments. OpenTSLM processes numerical time-series data alongside text inputs to deliver natural language explanations and answers. The model is designed to support applications such as IT support workflows by integrating technical signals with contextual information.
Key details
- OpenTSLM-Flamingo Gemma 3 4B has 5.24 billion parameters and requires about 10.5 GB of GPU memory in bfloat16.
- The model uses a Perceiver Resampler and cross-attention layers to combine time-series data with text.
- Requests can include up to four questions, each with one to 32 numerical series and up to 32,768 values per series.
- Real-time inference endpoints have a 60-second time limit and a 6 MB payload cap.
- Fine-tuning on Datadog’s ARFBench data improved accuracy from 5.0% to 45.9% on held-out questions.
Why it matters
AWS customers, including IT teams and companies using SageMaker, can now deploy a model that interprets both time-series and text data for real-time support and diagnostics. This enables more automated and context-aware responses in workflows such as IT incident resolution, potentially reducing manual intervention and response times. Organizations can also adapt the model to their own data and tasks, improving its relevance and accuracy for specific operational needs.
