Mira Murati unveils Inkling 2 Small: 594 billion parameters

Thinking Machines Lab, founded by Mira Murati, released the multimodal MoE model Inkling 2 Small on November 5, 2026. The model features 594 billion parameters (17 billion active), a context window of up to 1 million tokens, and a score of 1340 on DesignArena (Website). It is currently available via Tinker, with full weights under the Apache 2.0 license promised for next week following external safety checks.

Compiled by Sergey KostenchukPublished 2026-11-06Updated 2026-11-06

2026-11-06 Research OpenAI

🤖 Mira Murati unveils Inkling 2 Small: 594 billion parameters

Thinking Machines Lab released a multimodal MoE model on November 5: 594 billion weights, 17 billion active per token, a context window of up to 1 million tokens, and text and image inputs. On DesignArena (Website), it scored 1340, compared to 1333 for the closed GPT-6 Astra and 1328 for Kimi K3.

🌍 The company is closing the gap with closed frontier models in agentic coding and interface generation. The release strategy—initially hosted access via Tinker, followed by open weights under Apache 2.0 after external safety checks—sets a standard for large open MoE releases.

👤 You can try it now in Tinker (inference and fine-tuning), with full weights promised for next week. Note: "Small" does not mean "light"—594 billion weights will not fit in the memory of a standard computer, and the data processing terms for hosted fine-tuning have not been disclosed.

Source 1: https://thinkingmachines.ai/news/inkling-2-small/

Sources