Inkling 2 Small

Overview

Inkling 2 Small is a multimodal Mixture-of-Experts AI model from Thinking Machines Lab. It supports image inputs, variable effort, agentic coding, autoresearch, and fine-tuning through Tinker.

Inkling 2 Small is the second-generation model from Thinking Machines Lab. The website describes it as a Mixture-of-Experts model with 594B total parameters, 17B active, image input support, and variable effort. It is introduced as being available through Tinker, where users can try it in the Tinker Playground and use it for fine-tuning and sampling. The product is positioned around AI-native capabilities rather than conventional software workflows. The page emphasizes agentic coding, autoresearch, vision interpretation and creation, reasoning, and long-horizon autonomous work. Examples on the page include generating a city simulator, building a Linux PC emulator in Python with a browser interface, creating visual scenes in Blender, and generating a floor-plan walkthrough from a map. For coding and research workflows, the site says Inkling 2 Small was trained with reinforcement learning on long-horizon coding tasks across multiple agent harnesses, and that it is designed to work inside real agent loops. The page also describes an internal benchmark example where the model autonomously trained a forecasting model on Tinker, wrote its own evaluator, experimented with training recipes, and iterated on supervised fine-tuning and reinforcement learning. For multimodal use, the site says the model processes image inputs with a vision encoder and supports OCR, grounding, infographic interpretation, and visual reasoning. It also states that the model uses vision in the loop to interpret and validate visual coding output. Based on the first-party page, Inkling 2 Small appears to be a hosted AI model product accessible online through Tinker, with AI being the essential product itself.

Features

• Image input support
• Variable effort inference
• Agentic coding
• Autoresearch workflows
• Fine-tuning on Tinker
• Sampling on Tinker
• Vision interpretation
• Vision-based creation
• Long-horizon autonomous work
• Tinker Playground access

Tags

FAQ