Mira Murati's Thinking Machines Lab released Inkling, a 975-billion parameter open-weights model designed for enterprise use and fine-tuning rather than raw performance competition. The multimodal model leads U.S. open-source alternatives on the Artificial Analysis Intelligence Index but trails top Chinese open models on several benchmarks.
Pricing stands at $1.87 per million input tokens, positioning Inkling as an accessible option for developers building custom applications. Murati, who departed OpenAI as CTO in 2024, launched Thinking Machines to focus on practical AI deployment for enterprises seeking control over their model infrastructure.
The release reflects a strategic shift in how open-source AI vendors compete. Rather than chasing benchmark supremacy against proprietary models from OpenAI or Anthropic, Thinking Machines emphasizes Inkling as a foundation for customization. Organizations can fine-tune the model on proprietary data without depending on external API providers, addressing growing concerns about data privacy and operational independence.
The 975B parameter scale positions Inkling between typical consumer models and the largest frontier models. This sizing reflects real-world constraints. Training larger models demands resources few companies possess, while smaller models struggle with complex reasoning tasks. Inkling aims for the practical middle ground where enterprises need sufficient capability without the infrastructure costs of trillion-parameter systems.
Chinese open models' superior performance on certain tasks underscores ongoing competitive pressure from Beijing. Models like Alibaba's Qwen continue advancing rapidly, challenging Western assumptions about technological leadership in AI.
Thinking Machines competes in a crowded open-source landscape alongside Meta's Llama, Mistral, and others. Differentiation comes through Murati's credibility and the explicit focus on enterprise fine-tuning workflows. The company avoids positioning Inkling as a general-purpose replacement for frontier