[AINews] Thinky's Inkling: 975B-A41B multimodal, new best American Apache 2.0 open model (with Inkling-Small, 276B-A12B)
Thinking Machines released Inkling as an open-weights, Apache 2.0 multimodal model family built for customization rather than a SOTA claim.
The main model is described as a 975B-parameter MoE with 41B active parameters, trained on 45T tokens across text, images, audio, and video. It supports up to a 1M-token context in the open-weight checkpoints, with Tinker/API context described at 256K. The note says reviewers framed it as the strongest U.S. open-weight release so far, while still behind leading Chinese open-weight and closed models on some benchmarks. Inkling-Small is previewed at 276B total parameters and 12B active. Latent Space's note
The main model is described as a 975B-parameter MoE with 41B active parameters, trained on 45T tokens across text, images, audio, and video. It supports up to a 1M-token context in the open-weight checkpoints, with Tinker/API context described at 256K. The note says reviewers framed it as the strongest U.S. open-weight release so far, while still behind leading Chinese open-weight and closed models on some benchmarks. Inkling-Small is previewed at 276B total parameters and 12B active. Latent Space's note
score 8