On July 15, 2026, Thinking Machines Lab — the research company founded by former OpenAI CTO Mira Murati — released Inkling, a 975B-parameter Mixture-of-Experts model with 41B active parameters under Apache 2.0.
Key specs
| Inkling | |
|---|---|
| Total parameters | 975B |
| Active parameters | 41B (MoE) |
| Context | 256K tokens |
| Input modalities | Text, image, audio |
| License | Apache 2.0 |
| Weights | Full precision on Hugging Face |
| API pricing | ~$2 / $8 per 1M tokens |
What it is — and isn’t
Thinking Machines Lab explicitly positions Inkling as a fine-tuning base, not a frontier competitor. The model is designed to be adapted — domain specialists, vertical applications, enterprise customization — rather than to chase benchmark leaderboards.
That said, the specs are notable: 975B total parameters make it the second-largest open-weight model after Kimi K3 (2.8T), and unlike K3’s custom license, Inkling ships under Apache 2.0 with no commercial restrictions.
Why it matters
- US-origin Apache 2.0 at scale. The open-weight field has been dominated by Chinese labs (Kimi, DeepSeek, Qwen). Inkling is the first US-origin model above 500B with a genuinely permissive license — which matters for regulated industries and government contracts.
- Fine-tuning, not inference. By positioning as a base rather than a product, Thinking Machines Lab is betting that the value in open weights is downstream adaptation, not raw chat quality.
- Mira Murati’s first ship. Inkling is the first public model from the former OpenAI CTO’s new lab — a signal that the “post-OpenAI” generation of AI companies is starting to deliver.
Full specs: Inkling model page. For landscape context, see the 2026 LLM guide.