Altman calls Cerebras' 1200 token/s Astra model the biggest 'red pill moment' of his life
alexcovo_eth · x · 2026-10-22
Dave Blundin (MIT lab) shares that Harvard's Will Thompson got access to Cerebras' Astra model running at 1200 tokens per second — currently only Sam Altman and his top engineers have it, at roughly 10x the usual per-token price. The model is so fast that results appear before you finish typing a sentence. Altman calls it the biggest red pill moment of his life: once you've experienced it, you can never go back.
More from Infra
- PyTorch's TorchTitan adds DeepSeek-V4, Kimi K3 support, RL training and big scale gains — PyTorch · 2026-10-22
- Epoch AI tracks 95 AI data centers with 14.2 GW — still just 44% of global AI compute — haider1 · 2026-10-22
- Two CVSS 9.2 RCE Vulnerabilities in PHP Push a European Government Site Offline — HankYeomans · 2026-10-22
- Old H100s Now Rent Above Launch Prices as Compute-per-Watt Value Keeps Climbing — abhiadesai · 2026-10-22
- Jensen Huang on GPU residual values: newer is better, but old GPUs matter when everything sells out — AccBalanced · 2026-10-22
- Budget MoE workstation: 384GB DDR4 + one 32GB GPU is the sweet spot — Numerous-Ad6217 · 2026-10-22