Slow requests take up to 4 minutes on some occasions, feeling quite frustrated.
What works out more cost-effective, usage-based pricing or an API key?
I have the same issue and have activated usage-based pricing. However, I still get slow requests, and there is no way to pay for faster requests. Usage-based pricing only allows me to get more requests from GPT-4-turbo beyond the daily limit.
When using my own API key for Anthropic models, I run into rate limits when using Composer. It is also quite expensive, at about 10 to 20 cents per request with long context enabled.
The API key route is slightly cheaper (less than $0.01 per query), but probably not worth the hassle of managing your own key and dealing with rate limits. Usage-based pricing is more straightforward and gives you the same amount of fast requests
For faster responses, you might want to try claude-3.5-haiku - it’s about 3x cheaper than sonnet and usually pretty quick. The speed issues are likely due to high platform usage rather than which payment method you’re using
1 Like