The best thing I like is the latency. On my platform, I build voice agents for businesses, and for voice agents the most important factors are latency, time to first token, and tokens per second. So far, I haven’t faced any issues with latency, and the inference speed has been excellent. It’s still below Cerebras, but it works well for my use case. Another big plus is that we get newly launched models quickly. Review collected by and hosted on G2.com.

The API limits requests per minute, which is harming my business. I’d like to see the limits, or the TPM, increased even for lower-tier users. Some users don’t have a consistent need for higher limits, but others—like me—get traffic in bursts, so during peak times I run into 429 errors. Review collected by and hosted on G2.com.