For her first technological speech since leaving OpenAI in September 2024, Mira Murati strikes a big blow. The former technical director has just unveiled the first AI model of her start-up Thinking Machines Lab with a fully assumed ambition: to rely on open-source to redefine the sovereignty of corporate data.
In brief
- Thinking Machines releases Inkling, its first artificial intelligence model.
- Mira Murati, former CTO of OpenAI, heads this new laboratory specializing in AI.
- Inkling adopts an open source approach to encourage contributions from researchers and developers.
- The initiative stands out from the strategies favored by several major AI players.
A new AI breaking with the paradigm of closed models
Named Inkling, the AI model of Mira Murati is based on a Mixture-of-Experts (MoE) architecture. According to the official announcement from Thinking Machines, it has 975 billion parameters. However, it only activates 41 billion per text token processed. Result: artificial intelligence that is more efficient, faster and less expensive to run than a traditional massive model.
Designed as a customizable base via the Tinker adjustment tool, Inkling natively accepts textual, visual and audio data. It manages a context of up to 1 million tokens, or around 750,000 words. This revolutionary AI model was even pre-trained on 45 trillion tokens.
But its greatest particularity lies in its method of distribution. The complete weights are in fact published free of charge on Hugging Face under the Apache 2.0 license, without restriction of commercial use.
Decryption: Any company can download, host and locally modify theAI algorithm. A software decentralization approach that breaks with technical and economic dependence on proprietary APIs from OpenAI and Google!
In this context, Thinking Machines highlights an important point in its official press release : Inkling aims above all at research and experimentation. The AI model is therefore available to the community. In other words, researchers, developers and engineers can study it, improve it and design new applications. An approach that entirely reconnects with the principles of open source.
With Inkling, Mira Murati relaunches the AI infrastructure war
On MCP Atlas, a benchmark that measures how reliable an AI agent is in completing real-world tasks via the standard Model Context Protocol (MCP), the AI model scores 74.1%, compared to 44.7% for Nvidia’s Nemotron 3 Ultra. It is its main Western competitor in Open source AI.
On SWE-Bench Verified, which evaluates the ability of an agent to fix real software bugs on GitHub alone, it reached 77.6%. He also gains the advantage over Nemotron.
On Teminal Terminal Bench 2.1, Inkling is well ahead of GLM 5.2 from the Z.ai laboratory: 63.8% against 82.7%. Kimi K2.6 also dominates Humanity’s Last Exam, a doctoral-level test of scientific reasoning.
Thinking Machines takes the results head-on by asserting:
Inkling is not the most powerful model available today, open or closed.
Analysts, however, highlight a point less highlighted by the official press release: Inkling’s MoE design. It closely follows the architecture of DeepSeek-V3, an open source Chinese artificial intelligence model which shook up the market in 2025.
Important :
When an American laboratory supposed to offer an alternative to Chinese open source AI builds its first model on largely Chinese foundations, this only confirms the facts: best open source models on the market remain predominantly Chinese today.
Towards an economic recomposition of the artificial intelligence market?
Through Inkling, Mira Murati is not just adding an open-source option to the AI market. It also intends to disrupt the current dynamics of the sector. Especially since Inkling will have to compete with already well-established companies like OpenAI, Google and Anthropic.
Ultimately, two trajectories emerge.
- On the one hand, private and hybrid cloud giants could massively integrate Inkling to attract highly regulated sectors such as market finance and health.
- On the other hand, the pressure exerted by this high-level open source alternative could force closed system providers to drastically lower the cost of access to their own services to avoid the exodus of developers.
In any case, the global AI industry is entering a major strategic turning point. The battle for control of enterprise intelligence is no longer about server size alone. The freedom granted to developers now takes center stage.
Maximize your Tremplin.io experience with our ‘Read to Earn’ program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
