Together AI expands fine-tuning with GLM-5.3, Kimi K2.7-Code, live metrics, 30-70% price cuts
togethercompute · x · 2026-09-12
Together AI expanded its fine-tuning service with support for the latest open-weight models including GLM-5.3 and Kimi K2.7-Code (40+ models total).
Key updates:
- Live run metrics and in-flight experiment comparisons
- Early stopping when validation loss plateaus
- Dataset previews and validation before training
- Finer controls over training and data processing
- Prices down 30-70% on selected models
The goal is letting teams turn open-weight models into production-grade models quickly without handling each new architecture's technical challenges themselves.
More from Infra
- Viral Post: '"We Have the Compute" Was a Lie' — julian88888888 · 2026-09-12
- Open-source Deplo self-hosted deploy platform: no Docker, SSH or YAML, AI agents included — deplocloud · 2026-09-12
- Open-source models account for just ~3.2% of global AI training spend, analysis finds — 0xBekket · 2026-09-12
- OpenAI CFO: 'People are chasing GPUs — agentic AI is going to activate CPUs' — rohanpaul_ai · 2026-09-12
- 512GB DDR4 + 2x RTX 3090: What Local Models Should You Run on This Setup? — ArtifartX · 2026-09-12
- Meta Gives Every User a Free 2-vCPU VM With 100GB Storage, Compute Moat Even OpenAI Can't Match — signulll · 2026-09-12