If they ignore latency, they can do like 18K tokens/second/GPU of ≈Opus 4.7 "Flash", anon Just on two SuperPoDs they had by May, that's 295 million t…
If they ignore latency, they can do like 18K tokens/second/GPU of ≈Opus 4.7 "Flash", anonJust on two SuperPoDs they had by May, that's 295 million tokens/second.…