Which AI model for which task
You're burning money on the wrong models. Every team running multi-model AI operations makes the same mistake: they default to the most capable model for everything, then wonder why their API bill is $300/month when it should be $40. This guide fixes that.
Claude vs GPT vs Gemini vs local — the real answer isn't "pick one." It's knowing which task goes where. Creative synthesis and deep reasoning go to Opus. Code generation and structured extraction go to Sonnet. Mechanical tasks like classification, formatting, and templating go to Haiku. The wrong assignment on a high-volume task costs you $50/day. The right one costs $2.
The guide covers cost comparison with real production numbers from running 8 agents daily. You'll see what a 1,000-call research briefing actually costs on each Claude tier, what GPT-4o vs GPT-4o-mini looks like at scale, and when Gemini Flash is the answer nobody's using. The numbers will surprise you.
One of the most overlooked optimizations: flat-rate CLI subscriptions vs per-token API billing. For certain workloads — long iterative coding sessions, multi-step research chains — a $20/month Claude Pro subscription routed through an MCP server can replace $150/month in API tokens. This guide shows you the calculation. It's not always obvious, but it's always worth knowing.
Model routing by task type is the core framework. We've mapped 25 common agent task types to their optimal model — with cost per 1,000 calls and quality score for each. The decision tree for Opus vs Sonnet vs Haiku alone has saved production operations hundreds of dollars per month. Local models via Ollama get a full section: when latency matters more than quality, when you need air-gapped processing, and which models punch above their weight at the 7B and 13B parameter levels.
The included cost calculator is an actual spreadsheet with real formulas — plug in your task volumes and it outputs monthly cost across every major model option. No more guessing. The guide that saves you $50/day in wasted tokens pays for itself in the first hour.
Initial release. Model comparison matrix, routing guide, cost calculator, Ollama section, config templates.