PyyanThe AI Index

Pyyan / Compare / Atria Dawn Preview vs Inkling

Atria Dawn Preview vs Inkling

2 of 5

Open-Weight Models · verified 16 Sept 2026

SAtria Dawn PreviewShanghai AI Laboratorycurrent
TInklingThinking Machines Labcurrent
3 slots left
SpecificationAtria Dawn PreviewInkling
Summary744B agentic weights under the MIT licence, published without a launch.975B under Apache 2.0, shipped as a starting point rather than a finished product.
Context256K1M
Input ($/Mtok)Not publishedWeights only
Output ($/Mtok)Not publishedWeights only
LicenceMITApache 2.0
Max outputNot publishedNot published
Parameters744B total975B total, 41B active
Released14 Sep 202615 Jul 2026
Model IDinternlm/Atria-Dawn-PreviewNot published
CategoryOpen-Weight ModelsOpen-Weight Models
OfficialShanghai AI Laboratory ↗Thinking Machines Lab ↗

Highlighted rows are where these differ.

Atria Dawn Preview

  • Built on Zhipu's open GLM-5.2 rather than trained from scratch
  • 96.0 on DeepSearchQA and 86.5 on CyberGym, both reported by the lab
  • BF16 and FP8 checkpoints, served through SGLang or vLLM

Best for agentic work on your own cluster.

Full spec sheet →

Inkling

  • Pretrained on 45 trillion tokens of text, images, audio and video, with an encoder free architecture
  • Apache 2.0, weights on Hugging Face, same day fine tuning through the lab's Tinker platform
  • The lab earns from the compute customers use to adapt it, not from per token access

Best for fine tuning on your own data.

Full spec sheet →