Story perspectives
Open-Source 975B Inkling Model Matches Nvidia Benchmark
7/16/2026
1 of 3
Inkling Released
- Thinking Machines Lab released its open-weight model Inkling on July 15, 2026.
- Inkling has 975 billion parameters, activates about 41 billion per query, was trained on 45 trillion multimodal tokens, and is released under an Apache 2.0 license with weights on Hugging Face for fine-tuning via the Tinker platform.
- Thinking Machines says Inkling matches Nvidia’s Nemotron 3 Ultra on a coding benchmark while using about one-third the tokens, and a 276-billion-parameter Inkling-Small preview is available.
1 / 3
