Macaron AI's Mind Lab Sets New Benchmark with Trillion Parameter RL at 10% Cost, Now Integrated Into NVIDIA Megatron

What we drew from this source

The claims Via News extracted from this document. We point to the source; we don't replace it.

• Despite targeted adaptation, downstream evaluations on unseen benchmarks confirm that the model retains its broad general-purpose capabilities while gaining sharper task alignment

  80% confidence

• We don't blindly scale up, we scale smarter

  80% confidence

• Real intelligence learns from real experience

  80% confidence

• The next leap in AI won't come from a bigger data center, but from a fundamental breakthrough in how machines learn from experience

  80% confidence

• Slashed the time per RL training iteration by over 6x through a synchronized rollout and training architecture

  80% confidence

• Training on real user feedback can yield larger performance boosts than simply adding more pre-training data

  80% confidence

• LoRA-based reinforcement learning on Kimi K2 achieves the same alignment quality with just 10% of the GPU footprint required for full-parameter training

  80% confidence

• Training runs exhibit smooth, reliable learning curves with steadily increasing rewards and task success rates, free from instability or catastrophic collapse

  80% confidence

Cited in these Via News reports

• The Global AI Reckoning: Capability, Control, and the Widening Fault Lines of a Technology Race Without Borders →