Together AI Fine-tuning

Managed fine-tuning platform for open-source LLMs and vision models with LoRA, full fine-tuning, and RL support.

Paid· Usage-based; cost estimator in-product, no public price listFine-tuningMulti-model (any Hugging Face open-source model)

Visit website →

Best for

Pick Together AI Fine-tuning if you're a production team that wants to own custom open-weight models without buying a GPU cluster.

Skip if

Skip it if you're happy calling closed APIs, need a fixed published price, or just want a one-click hobbyist LoRA notebook.

Together AI Fine-tuning is a managed service for customizing open-source language and vision-language models on your own data without wrangling training clusters. It supports LoRA and full fine-tuning, reinforcement learning, tool-calling training on existing agent logs, and vision fine-tuning directly from PNG/JPEG/WEBP inputs. Under the hood it runs multi-GPU distributed training capable of handling 100B+ parameter models, with proprietary optimizations like UPipe for memory efficiency and extended context training (2-4x longer contexts at no extra cost).

It's aimed at production ML teams and enterprises who want to own their weights rather than rent a closed model. Any open-source model on Hugging Face Hub is fair game — DeepSeek, Qwen3, GLM, Gemma, Kimi K2, Llama-4 vision variants, and more. Pricing isn't fixed on the page but a cost estimator lets you preview training spend before launching a job, and fine-tuned models can be deployed straight onto Together's inference infrastructure.

Regional data residency (North America, Europe, Asia/Middle East), SOC 2 Type II, and ISO 27001:2022 certifications make it credible for regulated buyers. If you're stuck between fine-tuning on raw GPUs yourself and giving up on open weights entirely, this is the middle path.

Editor's take

Together has quietly become one of the most serious open-model training shops on the market, and the fine-tuning product reflects that — vision, RL, tool-calling and 100B+ scale are all first-class. The lack of transparent pricing is annoying, but if you're at the scale where you'd actually use it, the estimator and sales conversation are fine.

— The AI Tool Bible editorial team