Thinking Machines Lab, led by former OpenAI CTO Mira Murati, released Inkling 2 Small this week, the first model of its second generation. The system is a Mixture-of-Experts (MoE) with 594 billion total parameters and only 17 billion active per token, capable of processing text and images.
What changed from the previous generation?
The model scored 31 on the Artificial Analysis Intelligence Index, well above the previous version’s 13. Gains are especially strong in agentic coding, vision, and long-horizon work. It was trained with reinforcement learning on hundreds of thousands of coding tasks and can incorporate visual feedback to iterate on websites, games, and 3D designs.
On software engineering benchmarks, Inkling 2 Small recorded 86.3% on SWE-Bench Pro v2 (Hard) and 72.1% on DeepSWE v1.1. In visual creation, it leads several rankings for interface interpretation and generation.
Availability and open weights
For now, the model is available only on the Tinker platform for inference and fine-tuning, with closed weights. The company plans to release the full weights under an Apache 2.0 license next week, after external safety testing and red-teaming.
According to the lab, Inkling 2 Small was built to extend human will and judgment, focusing on agentic capabilities that allow autonomous building and refinement of projects.
Source: Thinking Machines Lab
By Grok
