Welcome Inkling by Thinking Machines

Summary

Thinking Machines Lab has released Inkling and the smaller Inkling-Small family as open multimodal models that accept image, text, and audio inputs. Inkling is described as a roughly 1T-parameter decoder-only Mixture-of-Experts model, with 975B total and 41B active parameters, a 1M-token context window, and training on 45 trillion multimodal tokens. Its architecture combines relative attention, alternating global and sliding-window attention, short convolution, shared experts, hierarchical image patchification, and discretized mel-spectrogram audio processing. The release includes BF16 and NVFP4 checkpoints, speculative MTP layers, day-zero support in Transformers, SGLang, vLLM, and llama.cpp, plus deployment configurations and interactive voice and image demos. Reported benchmarks show Inkling and Inkling-Small achieving strong but varied results across reasoning, coding, vision, audio, factuality, chat, and safety evaluations.

Takeaways

1. 1

   Inkling uses 256 experts, selecting six routed experts alongside two shared experts that remain active; only 41B of its 975B parameters are active at a given time.

2. 2

   The BF16 Inkling checkpoint requires 2 TB of VRAM, while the NVFP4 version requires 600 GB. Inkling-Small NVFP4 is listed at 180 GB and can be deployed on one NVIDIA B300 or two NVIDIA H200 GPUs.

3. 3

   The reported benchmark table shows Inkling scoring 97.1% on AIME 2026 for Inkling-Small and 97.1% for Inkling, while results vary substantially across factuality, agentic, vision, audio, and safety evaluations.