ProductHardwareLibrarySand to ThoughtCareersDocs
Book a callLog in

The research lab focused on inference.

TrainingAutomatic fine-tuning from your traffic.InferenceOpen-model inference, pay based on your latency requirement.

Models

  • DeepSeek V4.1 FlashDeepSeek V4.1 Flash
  • GLM-5.3 FlashGLM-5.3 Flash
  • KKev 9BKev 9B
NotesFrontier notes on inference, systems and hardware.MethodsHow we operate, and why.

Latest writing

  • Announcing Carat: An inference engine designed for Gemma 4Sep 2, 2026
  • ThesisJul 13, 2026
  • Affordable and open intelligenceSep 9, 2026
See all writing

Library

Notes & Methods

Announcing Carat: An inference engine designed for Gemma 4Designing Carat, an inference engine built specifically around the model architecture of Gemma 4.Sep 2, 2026
ThesisMost of the car is missing.Jul 13, 2026

Product

  • Training
  • Inference
  • Service tiers
  • Models and pricing

Developers

  • Quickstart
  • API reference
  • Errors
  • Status

Company

  • About
  • Thesis
  • Careers
  • Library

Legal

  • Privacy
  • Terms
  • SLA
  • Acceptable use
  • DPA
  • Security

Gradiated is a brand owned by PostHog.

© 2026 PostHog Inc.

Cambridge, --:--:-- GMT