Why Anthropic lags OpenAI on math: compute constraints and 400k GPUs coming online
haider1 · x · 2026-09-26
- The author wonders why Fable 5.1 and Opus 5.5 aren't cracking math problems as fast as OpenAI's models.
- Likely cause: Anthropic is more compute-constrained and can't throw ten thousand agents at each Millennium Prize Problem like OpenAI does.
- OpenAI has another 400k GPUs coming online; Anthropic's compute pipeline is unclear without relying on SpaceXAI.
More from Infra
- Cerebras built a 20x faster Grok bot: dinner booking done in 22 seconds — Sethwinterroth · 2026-09-27
- Firmware optimization cuts agent latencies up to 190x, now under 0.5ms — dee_hw · 2026-09-27
- Quail: open-source AI-SQL engine hits 1B+ input tokens/min on a single H100 — sh_reya · 2026-09-27
- Open-source AI-SQL engine Quail hits 1B+ input tokens/min on a single H100 — sh_reya · 2026-09-27
- Skypilot jokes about squeezing 25% utilization out of a $100M GPU fleet — skypilot_org · 2026-09-27
- Is local AI actually accessible? M5 Mac Studio wait times stretch to February — Marino4K · 2026-09-26