Inkling 2 Small

Inkling 2 Small is Thinking Machines Lab's second-generation MoE multimodal model (594B total / 17B active parameters), released 2026-11-05. It accepts text and image inputs (no audio) and generates text, with variable thinking effort. The model card lists a context window up to 1M tokens; Tinker serves 256K for serverless inference, and training and sampling are offered at 64K and 256K. Vendor-reported Artificial Analysis Intelligence Index v5.0 score 30.6. Flash-tier price/performance on Tinker for inference and fine-tuning; currently closed-weight with Apache 2.0 weights pending external safety testing. Official Tinker serverless inference (256K sampling-nvfp4 list): $0.30 / $0.03 cached / $1.00 per 1M input/cached/output tokens.