Thinking Machines Lab launches Inkling 2 Small model on Tinker
The startup says the model, trained on 72 Blackwell Ultra GPUs, beats Spark Open and Kimi K3 on three knowledge-work benchmarks.
Thinking Machines Lab unveiled a large language model called Inkling 2 Small on November 6, 2026, a model the company says can automate AI training tasks. It is available immediately on the company's cloud service Tinker, according to SiliconANGLE.
The model has 5,940 billion total parameters, with about 17 billion active per query. Users can set how deeply it computes per task, trading quality against cost efficiency, and can further train it on their own data for post-training customization.
Thinking Machines Lab trained Inkling 2 Small on a GB300 NVL72 rack cluster with 72 Blackwell Ultra GPUs. Pre-training took 68 hours and was followed by 9 days of reinforcement learning. The company also trained it on hundreds of thousands of complex, time-consuming coding tasks to strengthen programming and work automation.
The company says Inkling 2 Small outperformed open-source models including Spark Open and Kimi K3 on three knowledge-work benchmarks. It plans to release the model weights under an open-source license next week.
Reported funding round
Separately, Business Insider reported that Thinking Machines Lab, led by founder and former OpenAI CTO Mira Murati, is raising USD 5 billion in a round led by Axel at an expected valuation of USD 40 billion.
Quick answers
Where can I use Inkling 2 Small?
It is available immediately on Thinking Machines Lab's cloud service Tinker.
Will the model weights be released?
The company plans to release them under an open-source license next week.
How large is Inkling 2 Small?
It has 5,940 billion total parameters, with about 17 billion active per query.