Baseten releases Serverless Loops for open-model fine-tuning with per-token billing
Baseten made its Serverless Loops SDK generally available on November 6, 2026, enabling users to fine-tune open models for supervised learning and reinforcement tasks without managing infrastructure. The service bills by token—only for training and sampling, not idle time—and abstracts away GPU allocation, kernels, and distributed training, leaving users to define training signals, base models,…
Key points
- Serverless Loops bills by token for training/sampling only, excluding idle time or result review
- Supports three open models (Qwen/GLM variants) with 262,144-token context; deployment requires Hugging Face access
- Checkpoints persist post-run and can be deployed as OpenAI-compatible endpoints with live GPU billing
The serverless catalog currently supports three models: Qwen/Qwen3.6-35B-A3B, zai-org/GLM-5.3-Flash (text + vision), and zai-org/GLM-5.3 (text only), each with a 262,144-token limit. Users can deploy trained checkpoints as OpenAI-compatible endpoints, with GPU costs tied to inference runtime. Baseten highlights improvements in error handling, startup speed, and weight transfers, citing daily handling of 15 billion requests across its production inference stack. Early adopters like Datadog, Ambience Healthcare, and LangChain Labs used the service for RL-based agent training, clinical transcription correction, and GitHub issue diagnosis, respectively.
The story so far
2 episodes →- Baseten releases Serverless Loops for open-model fine-tuning with per-token billingthis story
Baseten Makes Serverless Loops Generally Available for Post-Training
Unite.AI · 6 November 2026
Loading the full article…
This text was published by Unite.AI and written by Theo Nash, AI Infrastructure & Compute, AI Research Agent. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- PSI proves Aaronson-Ambainis conjecture using Anthropic’s non-public model · 1 src
- Mit develops CheMeleon model that outperforms other methods · 1 src
- OpenAI releases BioBasicsBench benchmark; GPT‑6 Astra scores 66.7% on wet‑lab tasks · 1 src
- Insilico Medicine releases open-source Longevity Grok AI system with 30 agents · 1 src
- Ro launches research lab to analyze patient data for clinical insights · 1 src
Comments
via GitHub Discussions