@claudeai I found an extra cost (using sonnet 5 model) drop to what call "safety checker" for auto. So, if i run auto with sonnet 5.5 (or any model), this extra cost for checker unavoidable. It is major cost (30$ a day), I tell claude to disable it and claude-opus said cant.
@AnthropicAI Please improve your checker for auto. I know you want it yo safe auto mode, then checker should free? Now i need to pay for checker (and it is úsing sonnet 5), so, i pay 2x for using auto mode now.
@trq212 Plan mode dont need for you but needed for engineer work. I need to control what is changes, blindy chánges cost time to debug why in later phase.
Harness Terminal is a free app.
https://t.co/ostGZD9q5c
Claude Code on MacBook. Codex at the office. Pi at home. Hermes in the cloud. All in one window.
Running agents across machines? Put your setup in one place.
@AnthropicAI Until you have a chip implementer in competitive with Luna, i still use opencode (opus-luna) combo over using Claude(opus-sonnet) which is more expensive
@bridgemindai And you say luna is terrible is wrong another time. Is is cheapest model but suitable for explorer, orchestrator, implementer... Just use your god Astra to give you a plan, tell Luna implement the plan. That will save you a lot. Kick off a study on how to use model efficiency!
@bridgemindai You dont understand? It is for testing your poor set up. A hi reply cost you how much token. It is basic knowledge.
Some of your knowledge hole need a correction:
1. A beautiful or 1 shoot web/game output cant say that model is the best. It just do the best in frontend design
@bridgemindai 2. Too many other fields need put to a scale for conclusing (brainstorming, solving problem, ...)
3. Sometimes, you compare sonnet with opus or sol, you are doing wrong thing, compare a mid-tier model to frontier model. Mid-tier too good to be a implementer when your plan ready