Did Opus 5.5 really cost $10B? Analysts argue final training run is under 10% of that
teortaxesTex · x · 2026-11-11
Anthropic's claimed $10B spend on Opus 5.5 is drawing scrutiny: one calculation notes the company's cumulative training cost since inception through Q2 2026 was only $16B, making the single-model figure hard to believe.
teortaxesTex breaks it down:
- At 200B active params and 120T tokens, the final training run's compute at long-term rental rates is well under 10% of $10B — likely under 5%
- Data alone could easily have cost 4x the compute
In short, if the $10B number holds, most of it probably went to data and other costs, not the final training run.
Related event: Leak Claims Anthropic's Opus 5.5 Cost $10B, Drawing Skepticism(3 posts)→
More from Infra
- Baidu's ERNIE Evolve tops public bests on NanoGPT Speedrun, brings RSI to real inference optimization — xiaohu · 2026-11-11
- Baidu says RSI auto-optimized ERNIE inference, 6.24x decode speedup on Kunxin P800 — xiaohu · 2026-11-11
- Nvidia's supply commitments jump from $119B to $279B in one quarter, locking up 2027-28 memory — tengyanAI · 2026-11-11
- Train 27B models on a 24GB GPU: ternary optimizer cuts memory to 1 byte/weight — oguzeldereli · 2026-11-11
- NVIDIA expands Meta Muse partnership, invests $3B in Meta stock — armand_ruiz · 2026-11-11
- Neurons in a dish could be the newest computing machines for AI — JosephJacks_ · 2026-11-11