Introducing Inkling 2 Small - Thinking Machines Lab

thinkingmachines.ai24 min read

Thinking Machines Lab has introduced Inkling 2 Small, a Mixture-of-Experts model featuring 594 billion total parameters and 17 billion active parameters with image input and variable effort support. The model demonstrates significant performance gains across agentic coding, vision, knowledge work, and reasoning compared to previous generations. The company initially makes the model available through Tinker with closed weights, intending to release the full weights under an Apache 2.0 license following external safety testing. Open article →

Notable comments

Discussion is broadly optimistic about fast, domestically produced “open-enough” Mixture-of-Experts performance, while a few commenters question benchmark methodology, practical usability, and even branding/scope (closed weights first, wine comparisons, parameter naming).

HN · itkovian_ Claims training in nvfp4 was the key enabler: “training was entirely nvfp4 - this is best model by far that has ever been trained in that precision and explains why they were able to do it so fast.”

HN · a11r Highlights benchmark credibility: “refreshing to see benchmarks where the reporting team is not the clear winner,” noting Inkling’s claimed edge over GLM 5.3 Flash (if it survives outside the chart).

HN · ygouzerh Provides an informal but telling external validation: sent the same “wine testing” example to Claude Opus 5.5 medium; it failed while Inkling 2 Small succeeded.

More on this topic