Following her high-profile departure from OpenAI in September 2024, Mira Murati has finally pulled back the curtain on her latest venture. After two years of intensive development, her startup, Thinking Machines Lab, officially launched its flagship artificial intelligence model, Inkling, on July 15.<\/p>\n
\nFor some time, the open-source AI ecosystem has been heavily dominated by Eastern developers. Models such as Alibaba\u2019s Qwen, Z.ai\u2019s GLM, and Moonshot AI\u2019s Kimi have consistently outperformed Western alternatives on global leaderboards. While Nvidia\u2019s Nemotron previously served as the primary Western representative, it struggled to reach the “state-of-the-art” classification.<\/p>\n
Inkling marks a significant shift in this dynamic. As the most powerful open-source model built from the ground up by a Western laboratory, it aims to reclaim competitive ground. By releasing the full weights on Hugging Face under the permissive Apache 2.0 license, Thinking Machines Lab has ensured that the model is accessible without regional restrictions or proprietary gatekeeping.<\/p>\n
\nAt its core, Inkling utilizes a sophisticated Mixture-of-Experts (MoE) architecture. The model boasts a massive 975 billion total parameters, though it maintains efficiency by activating only 41 billion parameters during any single inference cycle. This design choice is intended to balance raw intelligence with operational speed.<\/p>\n
Key Launch Specifications:<\/strong> \nInkling is currently live on the OpenRouter platform, with pricing set at $1 per million input tokens and $4.05 per million output tokens. It is fully compatible with existing infrastructure, including Hermes and OpenClaw setups.<\/p>\n However, potential users should weigh these costs against current market realities. While Inkling represents a major milestone for Western open-source innovation, early performance data suggests that several competing models currently offer superior benchmark results at similar or even lower price points. For developers, the decision to integrate Inkling will likely depend on specific use-case requirements rather than raw cost-efficiency alone.<\/p>\n
\n* Release Date:<\/strong> July 15, 2025
\n* Architecture:<\/strong> Mixture-of-Experts (975B total \/ 41B active)
\n* Licensing:<\/strong> Apache 2.0 (Open Source)
\n* Accessibility:<\/strong> Available via OpenRouter<\/p>\nPractical Application and Market Positioning<\/h3>\n