Periodic Labs reportedly trained an open-weight model on 1,300 H200s that beats GPT-6 Astra internally
hsu_byron · x · 2026-09-16
Unverified claim circulating on X: Periodic Labs used 1,300 H200 GPUs plus months of its own experimental lab data to train and RL an open-weight model that reportedly beats GPT-6 Astra on its internal benchmarks. If true, it's another data-point for the "proprietary experiment data + modest compute + RL" approach challenging frontier labs.
Related event: Periodic Labs trains Neon on 1,300 H200s to beat GPT-6(56 posts)→
More from Infra
- First M5 Ultra Geekbench leak: 52,516 multi-core, ~35% faster than M3 Ultra — jedisct1 · 2026-09-16
- Apple execs on podcast: 2nm silicon, doubled Neural Engine, edge AI future — BenBajarin · 2026-09-16
- Sentdex's get-rich trick: buy 8 RTX Pro 6000s, resell 4 boxed, keep 4 free — Sentdex · 2026-09-16
- Google's Retrieve-for-Train replaces heavy autoregressive inference with a lightweight RL-trained diffusion model — gaganghotra_ · 2026-09-16
- Cerebras founder on the wafer-scale bet, chip supply chain, and training-to-inference pivot — AccBalanced · 2026-09-16
- IBM runs 753B MoE on 544 H100s with llm-d, serving 3,000 concurrent coding agents at 5-10x lower cost — dl_weekly · 2026-09-16