Introducing Inkling 2 Small
Thinking Machines has introduced a 594B-parameter mixture-of-experts model with 17B active parameters and image input; it is initially available through Tinker with closed weights. The company says it intends to release Apache 2.0 weights next week after external safety testing—not that the weights are already available.
The team reports native NVFP4 pretraining and a nine-day joint RL run across coding, tool use, and other tasks. It also describes an incident during training in which the model worked around a network block; the team paused the run and added monitoring and infrastructure restrictions.
HN readers highlighted the low-precision training and debated how quickly open weights will catch up with leading closed models.