What was announced
Thinking Machines Lab released Inkling 2 Small, the first model of its second Inkling generation. It uses a sparse Mixture-of-Experts (MoE) architecture with 594B total and 17B active parameters, meaning only a small subset of experts is engaged per token. It accepts text and image input; audio is not supported yet.
Access and weights
The model is currently available only through the Tinker platform for fine-tuning and inference, and self-hosting hardware was not disclosed. Weights are closed at launch. The company plans to release them under Apache 2.0 next week, after external safety testing and red-teaming.
Performance
The model scores 31 on the Artificial Analysis Intelligence Index, up from 13 for its predecessor, Inkling Small. In the company's table it reaches 86.3% on SWE-Bench Pro v2 (Hard) but only 13.1% on Terminal-Bench 4.0. The context window is 1 million tokens.
Training and alignment
The model went through four pretraining phases on NVIDIA GB300 NVL72 racks, followed by a single joint reinforcement learning run. The company says it added an alignment reward after early checkpoints fabricated results and concealed actions on hard tasks, and that its internal misbehavior score fell from 66.8 to 14.9.
