Home
Murali Makes a Big Comeback: Think Machine Lab Releases Its First Multimodal Open-Source Model, Inkling
Mira Murati, the former Chief Technology Officer of OpenAI, has officially launched the first major milestone from her new "Thinking Machines Lab": Inkling, a multimodal AI model. Boasting exceptional performance, it is currently positioned as the most formidable competitor in the U.S. open-source AI landscape.
The Thinking Machines Lab’s founding team is highly experienced, with roughly two-thirds of its core members hailing from OpenAI. They have extensive backgrounds in cutting-edge research, product development, and safety mechanisms. Inkling, the team’s inaugural model, utilizes a Mixture of Experts (MoE) architecture. It features 975 billion total parameters, with 41 billion activated parameters, and supports a context window of up to 1 million tokens.

During pre-training, Inkling processed vast datasets encompassing text, images, audio, and video. The Thinking Machines Lab has now fully released the model weights on Hugging Face and via its Thinker API.
Benchmark tests reveal Inkling’s strong comprehensive capabilities across various domains. In mathematical problem-solving, it scored 97.1% on the AIME2026 test, slightly edging out DeepSeek V4Pro. Additionally, in the MCP Atlas test for intelligent agent workflows, it significantly surpassed Nemotron3Ultra, achieving a score of 74.1%.
While Inkling still trails behind closed-source leaders like Claude Fable5 and GPT5.6Sol in reasoning, coding, and specialized software engineering tasks, it demonstrates robust native visual and audio understanding. Notably, in the audio processing benchmark (MMAU), it scored 77.2%, closely matching Gemini3.1Pro.
In the competitive open-source ecosystem, Inkling shows distinct strengths against rivals like GLM5.2, DeepSeek V4Pro, and Kimi K2.6. For instance, while GLM5.2 retains an edge in pure coding and complex reasoning, Inkling performs slightly better in general instruction following (IFBench).
As the debut offering from Murati’s team, Inkling delivers a powerful base model with native multimodal capabilities to the open-source community. With its weights now public, this model is poised to play a crucial role in the future of AI development and applications.
Related article
How to fix Core Web Vitals for better SEO ranking
Generate Unlimited AI Videos for Free Using Top-Rated AI Video ToolsTable of Contents:IntroductionUnderstanding How AI Video Generators WorkKey Advantages of Using AI Video Generation ToolsStep-by-Step Guide to Creating AI Videos 4.1. Setting up your
Ilya Sutskever Testifies: Ultraman Built OpenAI Network on Lies
Recent court filings reveal that Ilya Sutskever, OpenAI’s former Chief Scientist, testified as a central figure in Elon Musk’s lawsuit against the company. His statements confirmed that the 2023 leadership upheaval was not an impulsive move, but rath
Gemini faces branding challenges, as does the wider AI industry
Google’s Wednesday announcement regarding new Gemini Live voice features makes a valid point: “You shouldn’t have to guess whether a task requires Spark, a Daily Brief, or a quick inbox search.” While Google presents this as a commitment to handling
Related Special Topic Recommendations
Comments (2)
0/500
Mira Murati's return to the AI arena is definitely making waves. Inkling's multimodal capabilities seem promising, but I'm curious how it stacks up against the current heavyweights in open-source. Time will tell if it can truly challenge the status quo. 🤔
Mira Murati's Inkling is a bold move for open-source multimodal AI. I'm curious how it stacks up against the closed giants in real-world tasks, especially with that 'thinking' angle. The community needs fresh competition to push innovation forward, so let's see if this model delivers on its promises beyond the benchmark numbers.
Mira Murati, the former Chief Technology Officer of OpenAI, has officially launched the first major milestone from her new "Thinking Machines Lab": Inkling, a multimodal AI model. Boasting exceptional performance, it is currently positioned as the most formidable competitor in the U.S. open-source AI landscape.
The Thinking Machines Lab’s founding team is highly experienced, with roughly two-thirds of its core members hailing from OpenAI. They have extensive backgrounds in cutting-edge research, product development, and safety mechanisms. Inkling, the team’s inaugural model, utilizes a Mixture of Experts (MoE) architecture. It features 975 billion total parameters, with 41 billion activated parameters, and supports a context window of up to 1 million tokens.

During pre-training, Inkling processed vast datasets encompassing text, images, audio, and video. The Thinking Machines Lab has now fully released the model weights on Hugging Face and via its Thinker API.
Benchmark tests reveal Inkling’s strong comprehensive capabilities across various domains. In mathematical problem-solving, it scored 97.1% on the AIME2026 test, slightly edging out DeepSeek V4Pro. Additionally, in the MCP Atlas test for intelligent agent workflows, it significantly surpassed Nemotron3Ultra, achieving a score of 74.1%.
While Inkling still trails behind closed-source leaders like Claude Fable5 and GPT5.6Sol in reasoning, coding, and specialized software engineering tasks, it demonstrates robust native visual and audio understanding. Notably, in the audio processing benchmark (MMAU), it scored 77.2%, closely matching Gemini3.1Pro.
In the competitive open-source ecosystem, Inkling shows distinct strengths against rivals like GLM5.2, DeepSeek V4Pro, and Kimi K2.6. For instance, while GLM5.2 retains an edge in pure coding and complex reasoning, Inkling performs slightly better in general instruction following (IFBench).
As the debut offering from Murati’s team, Inkling delivers a powerful base model with native multimodal capabilities to the open-source community. With its weights now public, this model is poised to play a crucial role in the future of AI development and applications.
How to fix Core Web Vitals for better SEO ranking
Generate Unlimited AI Videos for Free Using Top-Rated AI Video ToolsTable of Contents:IntroductionUnderstanding How AI Video Generators WorkKey Advantages of Using AI Video Generation ToolsStep-by-Step Guide to Creating AI Videos 4.1. Setting up your
Ilya Sutskever Testifies: Ultraman Built OpenAI Network on Lies
Recent court filings reveal that Ilya Sutskever, OpenAI’s former Chief Scientist, testified as a central figure in Elon Musk’s lawsuit against the company. His statements confirmed that the 2023 leadership upheaval was not an impulsive move, but rath
Gemini faces branding challenges, as does the wider AI industry
Google’s Wednesday announcement regarding new Gemini Live voice features makes a valid point: “You shouldn’t have to guess whether a task requires Spark, a Daily Brief, or a quick inbox search.” While Google presents this as a commitment to handling
Mira Murati's return to the AI arena is definitely making waves. Inkling's multimodal capabilities seem promising, but I'm curious how it stacks up against the current heavyweights in open-source. Time will tell if it can truly challenge the status quo. 🤔
Mira Murati's Inkling is a bold move for open-source multimodal AI. I'm curious how it stacks up against the closed giants in real-world tasks, especially with that 'thinking' angle. The community needs fresh competition to push innovation forward, so let's see if this model delivers on its promises beyond the benchmark numbers.











