Inkling-Small matches or exceeds Inkling on reasoning and agentic tasks, the company said.
Nvidia-backed Thinking Machines Lab has unveiled a new open-weights model that performs “comparabl[y]” to Inkling, but at one-quarter of its size.
The company launched its first AI model Inkling earlier this month following a mega partnership with Nvidia that gave Thinking Machines access to GB300 NVL72 systems for training. The chipmaker also made a significant investment into the AI company.
Inkling-Small comes in at 276bn total parameters, with 12bn active. Compared to its bigger predecessor, the new model achieves similar performance with much less compute, said Thinking Machines. It matches or exceeds Inkling on reasoning and agentic tasks, the company added.
Artificial Analysis scores Inkling-Small 40pc on the Intelligence Index, placing it “well above average” relative to other comparable models. Inkling hits 41pc while DeepSeek V4 Flash matches Inkling-Small at 40pc. At 93 tokens per second, the model is also faster than the average, according to the benchmark.
On Humanity’s Last Exam, the new model scored above 31pc, ahead of Inkling’s 29.7pc, and on SWEBench-Verified, it crossed 80pc. The model takes text, image, and audio inputs, and outputs text.
Thinking Machines said that Inkling-Small, much like Inkling, is suited for real-world applications such as cropping, zooming and programmatic image inspection, improving usability on documents and charts where information could be difficult to read directly.
“Inkling-Small was made in pursuit of our mission to build AI that extends human will and judgement,” the company said. “Tinker customers have seen first-hand that the right fine-tuned model can outperform closed models on a variety of tasks, and do so faster and cheaper.”
The Mira Murati-led company is focused on creating highly customisable AI systems, according to its website.
Inkling is designed to be broad, the company explained at the time of its launch. “We trained it across agentic, reasoning, coding, instruction-following, factuality, vision and audio tasks, rather than narrowly optimising for one domain,” it said earlier this month.
Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.
You must be logged in to post a comment Login