← ClaudeAtlas

model-per-chatlisted

Load when someone runs many Claude Code / Claude desktop chats, agents, subagents or scheduled tasks and wants to decide which model each one should use (Fable, Opus, Sonnet, Haiku, or no model at all) to save plan limits or money without adding mistakes. Use it when a new model comes out ("Fable 5.2 just came out, who gets it?"), when limits run out too fast, or for "which model for each chat", "assign models to my agents", "save tokens across sessions", "qué modelo le pongo a cada chat", "a quién le pongo Fable", "Opus o Sonnet", "ahorrar límites", "se me acaba el plan", "que suba el esfuerzo solo", "Opus también es caro", "lo mejor pagando poco". It sets a cheap base (mid model, low effort) with a written rule to climb to the top model only for the hard or irreversible step and come back down. It reads the live plan meters and each session's current model, checks today's prices and plan rules on the official pages, scores every chat 1-10 by what a mistake there costs, maps the score and the run frequency t
RabbitDesigns22/claude-skills · ★ 0 · AI & Automation · score 75
Install: claude install-skill RabbitDesigns22/claude-skills
# Model per chat A fleet of chats all on the same top model burns the plan on the cheap work: a watcher that runs every 20 minutes to say "nothing new" costs the same per turn as the supervisor that approves client work. Moving everything to a cheap model fails the other way: the chats where a mistake costs money or a client start making them. One rule: **pay for intelligence where a mistake is expensive, and pay for volume nowhere.** ## 1. Read the facts live, never from memory Models, prices and plan rules change every few weeks. Before recommending anything, collect and write down with the time: - **Plan and meters now:** plan name, % used of each limit window and when it resets, per-model meters, and whether extra usage / usage credits are on and with what monthly cap. (Claude desktop: the usage card or the session-usage tool. Claude Code: `/status`.) - **Each chat's current model and effort.** Read it from the session metadata, not from what you assume. Someone may have already switched one. - **Today's model facts from the official pages** (the vendor's model announcement, pricing page, and the help-center article on how each model counts against the plan): which models the plan includes, which are billed outside the limits, any launch window and its exact end time with time zone, and the official advice on which model fits which work. Third-party posts are leads, not sources. If the official page doesn't state a fact, write "not stated" instead of filling the gap