Thinking Machines launches Inkling 2 Small

Thinking Machines Lab launched Inkling 2 Small on November 5, giving developers a new model for coding, visual interpretation and longer-running AI tasks. It is available now through the company's Tinker Playground and developer platform. The full model weights are not yet downloadable.

That distinction matters for organizations exploring custom AI assistants. A model available through a provider's platform can be tested and fine-tuned, but it is not the same as one a developer can download, modify and run independently. Thinking Machines says it plans to release the weights under an Apache 2.0 license the following week, contingent on finishing external safety testing and red-teaming. That is a plan, not an accomplished release.

The company's model card describes a model that takes text and images as input and produces text, with 594 billion total parameters but 17 billion active during processing and a context window of up to one million tokens. Thinking Machines reports improvements in coding and agent tasks over its previous model. Those benchmark and performance claims come from the company; they are not independent proof of real-world performance.

There are limits. The company warns of hallucinations, imperfect instruction-following and long-conversation errors, and advises human oversight in high-stakes uses. The model card does not establish how reliably or affordably it would work in a real small business.

Our read: The consequential change is another model that developers can put through real application tests today, with a possible future path to downloadable weights. Whether that path opens, and whether the model performs well enough for dependable customer-facing agents, remains to be demonstrated. A new capability is not evidence of higher business revenue or fewer jobs.