Inkling

Training compute

1.8×10²⁴ FLOP

Parameters

975B

Published

Jul 15, 2026

Inkling is an AI model developed by Thinking Machines (United States), first published in July 2026. It works in the language, multimodal, audio and 2 more domain, on tasks such as language modeling/generation and chat.

Training it took an estimated 1.8×10²⁴ FLOP of compute. The model has 975,000,000,000 parameters. It was trained on roughly 45T datapoints. Training ran on NVIDIA GB300 (Blackwell Ultra).

Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as likely.

Full record

Organization: Thinking Machines
Country of organization: United States
Domain: Language, Multimodal, Audio, Video, Vision
Task: Language modeling/generation, Chat
Training compute: 1.8×10²⁴ FLOP
Parameters: 975,000,000,000
Dataset size: 45T
Training hardware: NVIDIA GB300 (Blackwell Ultra)
Model accessibility: Open weights (unrestricted)
Open weights: Yes
Epoch confidence: Likely