Thinking Machines Raising $5 Billion at $40 Billion Valuation, BI Says
Thinking Machines is raising a $5 billion round led by Accel at a $40 billion valuation, Business Insider reported, as it released the Inkling 2 Small model.
Thinking Machines Lab, the artificial intelligence startup founded last year by former OpenAI chief technology officer Mira Murati, is raising a $5 billion funding round led by venture firm Accel, Business Insider reported. The round is expected to value the company at $40 billion, according to that report.
The fundraising news landed around the same time the company released a new large language model, Inkling 2 Small, which went live late Thursday, SiliconANGLE reported. The company has not formally announced the round’s terms.
What the reported round covers
Accel is leading the deal, according to Business Insider. The other investors, the mix of new and existing backers, and the expected closing date were not disclosed. A valuation is the price investors assign to the whole company when they buy in. Thinking Machines has not said how it would spend the money.
What Inkling 2 Small does
Inkling 2 Small is built to automate parts of AI development itself, according to SiliconANGLE. Thinking Machines trained it on GB300 NVL72 racks, systems that each hold 72 Blackwell Ultra graphics cards and half that number of central processors. The first stage of training, called pre-training, took 68 hours. The company then refined the model with a nine-day reinforcement learning process, a method in which the model improves by being rewarded for good results.
To sharpen its coding, the company gave the model hundreds of thousands of long programming tasks, some of them “autoresearch” exercises such as post-training neural networks and tuning kernels, the low-level code pieces that make a model run efficiently.
In one test, the company asked Inkling 2 Small to fine-tune Qwen3.5-4B, an open-source model, to make better predictions such as weather forecasts. Thinking Machines said its model doubled the forecasting score in 4.5 hours. The company also said Inkling 2 Small beat Spark Open, Kimi K3 and other open-source models on three widely used knowledge work benchmarks and scored higher on several sets of financial tasks.
Size, vision and availability
The model has 594 billion parameters, the internal settings a model learns during training. About 17 billion of them switch on when a user enters a prompt. A vision encoder that accounts for 292 million parameters lets it read images and check its own visual output, such as whether product pages on a website it builds display correctly. Developers can also set how much computing power the model spends on a task, trading quality against cost.
Inkling 2 Small is available through Thinking Machines’ Tinker cloud service. The company said it plans to release the model’s weights under an open-source license next week.
What it means for investors and customers
If the round closes on the reported terms, Accel and the other backers would hold stakes in a company priced at $40 billion while it is still private, so ordinary investors cannot buy shares. The company also says Inkling 2 Small learns faster than its earlier models during post-training, when customers adapt a model with their own data. The open release of the model’s weights next week would let outside companies run and customize Inkling 2 Small on their own hardware.
Source: SiliconANGLE
