AWS makes SageMaker AI Training Sessions generally available for custom model training
In brief
- SageMaker AI Training Sessions reached general availability at AWS on October 22, 2026.
- Developers write the training logic; AWS manages GPUs, scaling and distributed execution.
- Supported methods include supervised fine-tuning, reinforcement learning, custom loss functions and multi-turn RL.
- The service is live in N. Virginia, Oregon and Mumbai, priced by tokens processed.
What developers get
Training Sessions lets developers run their own post-training loops, according to Crypto Briefing's report on the launch. Supported approaches include supervised fine-tuning (SFT), reinforcement learning, custom loss functions and multi-turn RL. AWS handles GPU management, distributed execution and infrastructure scaling.
That's the whole pitch in two lines.
Access runs through Python SDK clients, either from a local machine or from SageMaker Studio. Developers can call session operations directly (including forward_backward and optim_step), so they aren't locked into a fixed training recipe. For those who'd rather not write every loop by hand, AWS also shipped a companion "SageMaker training sessions cookbook" with ready-made recipes for SFT and RL.
Models, regions and pricing
The supported open-weight models include Llama, Gemma, Qwen, Nemotron and gpt-oss. At launch, Training Sessions is available in US East (N. Virginia), US West (Oregon) and Asia Pacific (Mumbai), per the report.
On security, data is automatically encrypted at rest in S3. Server-side encryption can use customer-managed KMS keys, and access is governed through IAM managed policies.
Pricing is usage-based and tied to the volume of tokens processed.
Where it fits in SageMaker
Training Sessions builds on earlier serverless customization features inside SageMaker. Serverless full fine-tuning arrived in August 2026, Crypto Briefing said, and support for custom sequence lengths followed on October 21, 2026 (one day before this launch).
The outlet drew a clear line between the two products. In its view, full fine-tuning covered teams that wanted a more packaged experience, while Training Sessions targets developers who want to write custom loss functions and control optimizer steps without owning a cluster.
Frequently asked questions
What is Amazon SageMaker AI Training Sessions?
It's a serverless SageMaker feature for customizing AI models. AWS announced general availability on October 22, 2026, according to Crypto Briefing. Developers write the training logic while AWS manages GPUs, distributed execution and scaling.
Which training methods and models does Training Sessions support?
Supported approaches include supervised fine-tuning, reinforcement learning, custom loss functions and multi-turn RL. Supported open-weight models include Llama, Gemma, Qwen, Nemotron and gpt-oss, per Crypto Briefing.
Where is SageMaker Training Sessions available and how is it priced?
At launch it's available in US East (N. Virginia), US West (Oregon) and Asia Pacific (Mumbai). Pricing is usage-based and tied to the volume of tokens processed, according to the report.
How does Training Sessions differ from SageMaker serverless full fine-tuning?
Crypto Briefing said full fine-tuning, which arrived in August 2026, served teams wanting a more packaged experience. Training Sessions targets developers who want to write custom loss functions and control optimizer steps without owning a cluster.
