Former OpenAI CTO Mulati created a multi-modal Inkling model, which is known as the strongest open source AI in the United States. The Thinking Machine Laboratory founded by former OpenAI CTO Mira Mulati launched the first multi-modal AI model Inkling trained from scratch, which is considered by the outside world to be the strongest open source model in the United States.

The multi-modal Inkling model created by former OpenAI CTO Mulati is unveiled, known as the strongest open source AI in the United States. The Thinking Machine Laboratory, founded by former OpenAI CTO Mira Mulati, has launched the first multi-modal AI model Inkling trained from scratch. It is considered by the outside world to be the strongest open source model in the United States. The model adopts a hybrid expert architecture, with a total parameter of 975B, an activation parameter of 41B, and a longest context of 1 million tokens. The pre-training covers 45 trillion Tokens data, covering four modalities: text, image, audio and video. The weights are currently open for download. The team has a luxurious lineup, but reasoning programming is still inferior to China's open source powerhouses. About two-thirds of the core members of the Thinking Machine Laboratory are from OpenAI. The team has led cutting-edge research, product and security work at the organization, and the lineup is extremely luxurious. Yet even so, Inkling still lags behind the Chinese open source model in inference and programming. GLM 5. 2 leads the way in coding, agent reasoning and complex reasoning tasks, with SWEBench Pro scoring 62.1% vs. 54.3%, and Terminal Bench 2. 1 scoring 82.7% vs. 63.8%. DeepSeek V4 Pro maintains its advantage in coding and factual tests, and Kimi K2. 6 also wins in multiple technical benchmarks. However, Inkling successfully surpassed DeepSeek in the field of mathematics, and achieved a high score of 97.1% in the AIME2026 test. The multi-modal capabilities are outstanding, but there is still a gap against the closed-source giants. Compared with its domestic open source competitors in the United States, Inkling has obvious advantages and comprehensively suppresses Nvidia's Nemotron 3 Ultra in terms of reasoning, coding and agent workflow. Facing closed source giants such as Claude Fable 5, GPT-5.6 Sol and Gemini 3.1 Pro, Inkling still has a clear gap in extreme reasoning and software engineering autonomy. Claude Fable5 reaches 95.0% on SWEBench Verified, and GPT-5.6 Sol reaches Terminal Bench 2.1. It scored 89.5 points, far exceeding Inkling. But in terms of multi-modality, Inkling shows good competitiveness, with a score of 73.3% in the visual benchmark MMMU Pro and a score of 77.2% in the audio processing MMAU test. The gap with the 82.5% of Gemini 3.1 Pro is not big. via AI News (author: AI Base)