Thinking Machines Labs has released Inkling 2 Small, a model designed to automate tasks including coding, AI research and business work. It is available through the company’s Tinker cloud service. The company says it plans to release the model’s weights under an open-source license the following week, which would let developers run or adapt the model themselves.
The launch comes as the company is reportedly seeking a large funding round. Business Insider reported that Thinking Machines is raising $5 billion in a round led by Accel, at a potential $40 billion valuation. Those fundraising details are reported, not confirmed by Thinking Machines in the information provided. The company was founded last year by Mira Murati, OpenAI’s former chief technology officer.
What can the model do?
Thinking Machines trained Inkling 2 Small on a cluster of GB300 NVL72 systems, each with 72 Blackwell Ultra graphics cards. The initial pre-training took 68 hours, followed by a nine-day reinforcement-learning process. The company says it trained the model on hundreds of thousands of extended coding challenges, including exercises involving neural-network training and optimization of AI software components.
In one test, the company asked Inkling 2 Small to fine-tune Qwen3.5-4B to improve its forecasting ability. Thinking Machines says the smaller model doubled Qwen’s forecasting score in 4.5 hours. The company also says Inkling 2 Small outscored Spark Open, Kimi K3 and other open-source models on three knowledge-work benchmarks, and scored higher on several financial-task datasets. It did not identify the benchmarks or provide test settings in the announcement, so the results are company-reported rather than independently verified here.
How is it built and accessed?
Inkling 2 Small has 594 billion parameters, the adjustable values a model uses to produce answers. About 17 billion are active for a prompt, and the model can vary how much computing power it uses. A vision encoder, with 292 million parameters, lets it analyze images and inspect the appearance of graphics it generates. For example, Thinking Machines says it can check whether an online store’s product pages display correctly as well as review the code.
Developers can also customize the model through post-training on their own data. Thinking Machines says Inkling 2 Small learns faster than its earlier models, which it says could make this customization more cost-efficient. The company has not provided pricing details in the information available.
Why it matters
Developers can access Inkling 2 Small through Tinker and use it for coding, image checks and other tasks, while a planned release of its weights could allow more customization. The company’s performance claims have not been independently verified here, and pricing and the exact terms of the planned weights release are not specified.
