Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released its first in-house foundation model on July 15: Inkling, an open-weight multimodal system available immediately on Hugging Face and through the company's Tinker fine-tuning platform.

What shipped

Inkling is a mixture-of-experts model with 975 billion total parameters, of which 41 billion are active for any given task. It was trained from scratch on 45 trillion tokens of text, images, audio and video using Nvidia GB300 NVL72 hardware, and it natively reasons across text, image and audio inputs in a single model rather than bolting on separate encoders. It supports a context window of up to 1 million tokens and lets developers tune how much "thinking effort" the model spends per query to balance cost against quality. The full weights are released under the permissive Apache 2.0 license, meaning companies and developers can download, modify and redeploy the model with no usage restrictions. Alongside the flagship, Thinking Machines shared a preview of Inkling-Small, a 12-billion-active-parameter variant aimed at lower cost and latency.

Where it stands

Thinking Machines has been explicit that Inkling is not positioned to beat the strongest closed models from OpenAI, Anthropic or Google on raw capability. Instead, the company is pitching a combination of multimodal support, tunable inference cost, and direct compatibility with Tinker for fine-tuning as the reason enterprises would choose an open base model over a closed one. It is the company's first shipped model since it was founded by Murati in 2025 with several other former OpenAI and Google researchers, and follows the earlier release of Tinker, its API for fine-tuning open models on custom infrastructure.

Why it matters

Inkling adds a well-funded, US-based entrant to a field of large open-weight releases this year that has otherwise been dominated by Chinese labs such as Moonshot AI and DeepSeek. By open-sourcing the full weights under Apache 2.0 rather than a more restrictive license, Thinking Machines is directly courting the same enterprises and developers who fine-tune open models for regulated or cost-sensitive deployments — a segment where AgentsAI has tracked steadily growing agent-tooling demand as companies look to run customized models instead of routing every task through a frontier API.