We’re glad to share that Thinking Machines has awarded Kōzu AI a Tinker Research Grant: $5,000 in compute credits on Tinker, their platform for fine-tuning and reinforcement-learning workloads.
For a small lab this is the best kind of news. It isn’t a headline, it’s runway. Credits are hours, and hours are experiments we can now run instead of queue.
Why it fits
The grant lines up with our stack almost exactly. Stellarator, our orchestration layer, was already built as a control plane for Tinker, so none of these credits go to setup. They go into the loop we run every day: distill, fine-tune, evaluate, repeat, across the line from Deimos through Europa to Ganymede.
More room in that loop means more of what we care about: tighter reasoning traces, shorter chains of thought, and models that spend their token budget on purpose.
Where it points
This grant will speed up our work toward models whose reasoning is earned and not assumed: more intelligence from fewer tokens, proven on tasks that matter.
Thank you to the team at Thinking Machines. We’re looking forward to seeing what we can make with the extra time.
