Thinking Machines Lab Launches Inkling 2 Small With Weights Closed
The 594-billion-parameter model is available on Tinker for inference and fine-tuning, with a full Apache 2.0 weight release planned after safety testing.

Thinking Machines Lab launched Inkling 2 Small on Thursday, Nov. 5, with hosted access through Tinker and no downloadable weights at launch. Developers can use the model there for inference and fine-tuning, while its listed license is Tinker Terms of Use. The company plans to release the full weights under an Apache 2.0 license after external safety testing and red-teaming, with timing described only as “next week.”
A July staged-access framework presented Inkling and Inkling-Small as open-weight models and tied broader availability to safety evidence and ecosystem readiness. “On a safe path to openness, each stage should widen access only when the evidence supports it,” Thinking Machines wrote. The launch begins with hosted model use, while the full-weight release remains a later step with no specific calendar date.
Inkling 2 Small has 594 billion total parameters, with 17 billion active per token. It uses a sparse mixture-of-experts architecture that routes each token to eight of 384 experts. The total parameter count describes the model’s overall size; the active count and expert routing describe its per-token configuration. The model accepts text and image inputs, offers a context window of up to one million tokens and lets users adjust reasoning effort. The model is intended for coding, vision, knowledge work and reasoning.
The Artificial Analysis Intelligence Index combines nine evaluations across knowledge work, coding, science and general knowledge. Inkling 2 Small’s launch figure was 30.6 under version 5.0, while the current listing shows 31 under version 5.0.1. The two figures use different version labels, and the index score combines results across several test areas rather than reflecting a single task.
Tinker lists prices of $0.30 per million input tokens and $1 per million output tokens. The pricing table is dated Oct. 19, 2026, before the model’s Nov. 5 release. The company has not given a more precise date or time for the planned Apache 2.0 weight release than “next week.”