Last updated: November 6, 2026.
Thinking Machines Lab introduced Inkling 2 Small on November 5, 2026. Despite the Small name, it is a Mixture-of-Experts model with 594 billion total parameters, 17 billion active parameters, native image understanding and a context window of up to one million tokens.
Quick answer: Inkling 2 Small is a 594B-parameter multimodal MoE model. It activates 17B parameters per token, accepts text and images, supports up to 1M tokens of context and is available now in Tinker Playground.
Inkling 2 Small specs
| Specification | Inkling 2 Small |
|---|---|
| Total parameters | 594B |
| Active parameters | 17B |
| Architecture | Sparse Mixture-of-Experts |
| Layers | 48 |
| Experts | 384 total, 8 routed per token |
| Input | Text and images |
| Context | Up to 1M tokens |
What can Inkling 2 Small do?
The model is designed for agentic systems, tool use, coding assistants, chatbots, retrieval-augmented generation, multilingual work and visual understanding. Thinking Machines says the second generation improves agentic coding, vision, knowledge work and reasoning.
How good is it?
Thinking Machines reports an Artificial Analysis Intelligence Index score of 31. The company attributes the improvement to more efficient pre-training and a large unified reinforcement-learning run. Independent testing will still be important as the model reaches more users.
How to try Inkling 2 Small
Inkling 2 Small is available through Tinker Playground. Developers can also use the model for sampling and fine-tuning through Tinker.
The weights are initially closed. Thinking Machines says it intends to release full weights under the Apache 2.0 license after external safety testing and red-teaming.
Why the 1M context window matters
A one-million-token context window is relevant for large code repositories, long documents, agent traces and retrieval workflows that need to keep large amounts of information active in one session.
Why use a Mixture-of-Experts architecture?
Instead of activating all 594B parameters for every token, Inkling 2 Small routes each token through a small subset of experts. That gives the model a large total capacity while keeping active compute much lower than the headline parameter count suggests.
FAQ
Is Inkling 2 Small open source?
Not at launch. Thinking Machines says full weights are planned under Apache 2.0 after safety testing.
How many parameters are active?
17 billion active parameters out of 594 billion total.
Does it support images?
Yes. It accepts text and image inputs and generates text output.
Sources
Explore more: AI Models – Latest AI Tools
