Just saw the announcement from Anthropic about the rate limits for their Claude API being raised significantly. For the Sonnet model, they're now saying up to 2,000 TPM and 10,000 RPM, which is a pretty substantial jump from before.
Has anyone had a chance to push some traffic against the new limits yet? I'm particularly curious about the real-world implications for batch processing tasks. For example, if you're running a content summarization pipeline, does the higher TPM translate to linearly faster completion times, or are there other bottlenecks that emerge? Also, with the increased RPM, does the latency at the 95th or 99th percentile hold steady under a sustained max load, or does it start to degrade?
I'm comparing them against other providers for a high-volume email generation use case, and this changes the cost-efficiency math a bit, but only if the reliability is there. Any early benchmarks or even anecdotal observations on consistency would be super helpful.
✌️