Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has released its first major language model under an enterprise-friendly Apache 2.0 open source license. The model, called Inkling, is a natively multimodal mixture-of-experts system with 975 billion total parameters capable of reasoning across text, images, and audio.
Inkling was designed to balance cost against performance through a novel controllable thinking effort mechanism that allows developers to adjust how much computational effort the model applies to each task. This setting can be tuned from 0.2 to 0.99, enabling users to optimize for speed or accuracy depending on their specific use case.
On key benchmarks, Inkling scored 77.6 percent on SWE-Bench Verified, outperforming fellow US open rival Nvidia Nemotron 3 Ultra 71.9 percent. It also achieved 91.4 percent on VoiceBench, competitive with leading proprietary models. However, it trails top Chinese open-weights models like DeepSeek V4 Pro on certain coding benchmarks.
The release is significant for Western developers seeking an open-weight alternative to Chinese models. The open-weight landscape has been dominated by Chinese developers, with DeepSeek, GLM, and MiniMax producing leading models. Inkling provides a US-developed alternative that offers competitive performance with more familiar licensing terms.
Alongside the flagship model, Thinking Machines announced a preview of Inkling-Small, a lighter 276-billion-parameter alternative optimized for latency-sensitive and cost-constrained workloads. The models weights are available on Hugging Face and through the companys Tinker API platform.
This article was adapted from VentureBeat. Read the original here.
