@jake @kstars111 Thanks for the points about tool calls. I’ll add this to the docs today, but to summarise, a tool call is any action the AI decides to take outside of writing it’s own output. This does include reading and editing files, searching the web, and grepping code. Terminal commands only count if the Agent runs them inline, as it monitors the output itself - if it just suggest a terminal command, this is not a tool call!
@JonoGee Great to hear things have been working well for you recently! We’ve been working hard to resolve the capacity issues, mainly by distributing the load across multiple Claude hosters, such as Anthropic themselves, and AWS Bedrock. As you’ve mentioned, we should be at a much better place now than we were a week ago!
@john-savepoint I don’t believe we have any ongoing issues with being charged twice as you describe. I know we’ve spoken already on this, and I’ll chase up your email now and get this sorted for you!
@ailee As described in the initial post, Claude 3.7 is built for a different way of working to Claude 3.5 - it is not simply just an upgrade on the last model. Claude 3.7 works best in more expensive environments (like Claude Code!), and while we have worked hard to make its performance comparative to our other models, we wanted to ensure we didn’t lock away any of Claude 3.7’s ability just because it works differently to our other models.
As Jono says above, a lot of users are now happy with Claude 3.7’s performance, and we are, of course, still tweaking and refining it to try and iron out any weird behaviors. However, Max mode for Claude runs at a cost that could not be included in the Pro plan, so there was never an option to include such a setup in the base plan, irrespective of whether we called it “Max mode for Claude”, or just claude-3.7-thinking.