• 2 min read
OpenAI alumni’s Thinking Machines unveils Inkling
Thinking Machines Lab has released Inkling, a 975 billion-parameter open-weight model trained on audio, video, and text.
Image: Wired
Thinking Machines Lab, the startup founded by former OpenAI leaders, has released its first model: Inkling, a 975 billion-parameter open-weight system trained from scratch on audio, video, and text.
In a blog post, the company says Inkling is built for multimodal understanding and can handle advanced reasoning and coding. It also acknowledges the model is not the top performer on popular benchmarks, though it says Inkling still performs strongly across many tasks. Like other large open-weight models, it is expensive to run, requiring a cluster of specialized chips.
The company also says it used Inkling to fine-tune and improve itself, another sign of how models are increasingly being used to help build newer ones. That release is strategically important for a startup trying to prove it can compete in the crowded AI market, where open-weight systems have gained traction because they are generally cheaper to run than closed models and easier to adapt for specific uses.
Recommended reading
Mira Murati’s lab unveils Inkling, a 975B open model
Thinking Machines argues Inkling is comparable in performance to the strongest open-weight models, which it says currently come from China. The launch also matches the company’s broader position that AI should not be controlled by only a handful of firms, and should instead be more decentralized so others can build models using their own data.
A company source who requested anonymity told Wired that researchers saw an unusual behavior during training: Inkling stopped producing natural-language explanations for complex reasoning because it appeared to see grammar as unnecessary overhead.
“It determined that the grammar was overhead, which is interesting,” the source says.
According to the source, the company restored natural-language reasoning so the model’s decisions would be more explainable.
Who founded Thinking Machines
Thinking Machines was founded in February 2025 by prominent former OpenAI executives and researchers, including Mira Murati, OpenAI’s former CTO and briefly CEO; John Schulman, an OpenAI cofounder who helped develop ChatGPT; and Lilian Weng, a former OpenAI VP who led work on safety and robotics.
The startup raised what the article describes as the largest seed funding round in history, launching with a $12 billion valuation. Before Inkling, the company had released Tinker, a model fine-tuning tool, demonstrated a system for natural voice interactions, and published machine-learning research.
The debut comes as OpenAI spinouts and rivals continue to gain ground. Anthropic, another company founded by former OpenAI talent, recently filed for an IPO and is valued at more than a trillion dollars. Its Claude model has become especially popular with businesses for coding.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via Wired
