I was quite excited to used it, but after burning 30$ in just a 10 minute workflow of analysis + refactor, I don't see a way in the short future of using this from a country where dollar is not the first currency.
Thanks for sharing. Yes one thing is - if we "keep the cost the same" but newer models do more and more reasoning, they may end up costing _more_ in the end. Which defeats the purpose of course.
There are definitely some issues with this new release. Sonnet 4.5 is even more token hungry than Sonnet 4 and thus even more expensive per task. While I'm sure prompt engineering from both Anthropic and Kilo will help, right now, Sonnet 4.5 is doing far worse at code refactoring and debugging than Sonnet 4 just a few days ago. I am still using Sonnet 4, but the performance has substantially degraded. I am going to stay away from both of these models for awhile until the performance problems are fixed.
I ran 1 request, to build a plan (!) to refactor an 800 loc file, running from Architect mode. it proceeded to read 5 more files (about 3.5K loc), and that resulted in 300K input tokens (0 cache hits) at a price of $1.29. i believe if I would have run it in Orchestrator mode with write permissions, I could've burnt a GDP of Moldova with a single request
I was quite excited to used it, but after burning 30$ in just a 10 minute workflow of analysis + refactor, I don't see a way in the short future of using this from a country where dollar is not the first currency.
Thanks for sharing. Yes one thing is - if we "keep the cost the same" but newer models do more and more reasoning, they may end up costing _more_ in the end. Which defeats the purpose of course.
There are definitely some issues with this new release. Sonnet 4.5 is even more token hungry than Sonnet 4 and thus even more expensive per task. While I'm sure prompt engineering from both Anthropic and Kilo will help, right now, Sonnet 4.5 is doing far worse at code refactoring and debugging than Sonnet 4 just a few days ago. I am still using Sonnet 4, but the performance has substantially degraded. I am going to stay away from both of these models for awhile until the performance problems are fixed.
I ran 1 request, to build a plan (!) to refactor an 800 loc file, running from Architect mode. it proceeded to read 5 more files (about 3.5K loc), and that resulted in 300K input tokens (0 cache hits) at a price of $1.29. i believe if I would have run it in Orchestrator mode with write permissions, I could've burnt a GDP of Moldova with a single request
https://open.substack.com/pub/hamtechautomation/p/claude-sonnet-45-the-ai-model-that?r=64j4y5&utm_medium=ios