THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

AI / NEWS BRIEF

Thinking Machines launches Inkling 2 Small on Tinker

The 594-billion-parameter multimodal mixture-of-experts model is available for inference and fine-tuning, while its promised Apache 2.0 weights remain pending safety work.

By AIQuantumScore · · Event 2026-11-05

A second-generation model with sparse activation

Thinking Machines Lab has released Inkling 2 Small through its Tinker platform and Playground. The model card describes a 48-layer multimodal mixture-of-experts transformer with 594 billion total parameters and 17 billion active per token. It accepts text and image inputs, produces text and supports context windows up to one million tokens. Tinker currently provides inference and fine-tuning access.

The weights are not open yet

Inkling 2 Small launched with closed weights under Tinker's terms. Thinking Machines says it intends to release the full checkpoint under Apache 2.0 after external safety testing and red-teaming, but that is a future commitment rather than current access. Independent model ledgers confirmed the Nov. 5 Tinker release and continue to classify the model as proprietary until the files actually appear.

Independent testing shows progress with important gaps

OUR ANALYSIS

Artificial Analysis gives Inkling 2 Small an Intelligence Index score of 31, up from 13 for Inkling Small, but also reports a higher cost and longer time per index task. Other independent benchmark coverage remains incomplete, including several coding, cyber and legal evaluations. The release is significant as a substantial new trainable multimodal model, but broad superiority is not established and the promised open-weight milestone should be assessed only when downloadable weights and licence files are public.

More to follow