Together AI has expanded its fine-tuning service to support a wider range of open-weight models, including GLM-5.3 and Kimi K2.7, while introducing live experiment tracking and finer controls over the training process.

  • New model support includes GLM-5.3, Kimi K2.7, Qwen 3.8-27B, Gemma 4, and DeepSeek-V4 variants.
  • Live metrics are now exposed via API, CLI, and UI, allowing users to monitor loss and gradient norms in real-time.
  • Expert LoRA allows adapters to target expert layers in Mixture-of-Experts models, improving recall of new knowledge to 89% compared to 15% for standard adapters.
  • Early stopping automatically halts runs when validation loss plateaus and refunds unused training steps.
  • Arbitrary batch sizes are supported via gradient accumulation to handle sequences that exceed GPU memory.
  • Prices for training LoRA adapters have been reduced by 30% to 70%, with specific savings for models like the gpt-oss series.
  • Pre-flight file validation checks JSONL datasets for schema errors before job submission, and sample weights allow per-example influence control.

These updates aim to reduce technical friction and costs while enabling more precise control over model development workflows.