I run OpenCode with a different model for every job, and I haven't hit a rate limit yet

Published Oct 17, 2026, 7:00 PM EDT Nolen began their writing career in 2019, with three years dedicated to editing the Creative section at MakeUseOf. Their expertise lies at the crossroads of technology and creativity, covering areas like photography, video editing, and graphic design. Outside of work, you'll often find Nolen diving into a good book, writing their own stories, or playing video games. If you use Claude daily or for heavy projects, you've probably watched the usage meter as closely as the actual output. Anthropic has raised the limits a few times this year, which does help, though there's still a weekly cap behind all of it, and Opus chews through that faster than the lighter models. I think what stings is how much of the allowance goes to tasks that never needed a top model in the first place, like asking what one setting in a config file does. I've been using OpenCode as a cheaper alternative for a minute now, and I've mainly been treating it as a backup for when Claude runs dry. But swapping to a free model when a paid one runs out is still leaning on one usage bucket at a time. So the plan this time was to stop picking out of habit and be more intentional with handing out jobs to different models instead. The problem with leaning on one model for everything Every tiny request went to the heaviest model I have Claude is my main AI tool, but the gist of this applies to all the other premium agents too. Claude splits its limits two ways. There's a rolling five-hour session window, plus a weekly cap on top of that, and on Pro that weekly cap covers every model, with the chat app and Claude Code pulling from the same pool. I must say, the five-hour side has gotten better throughout the year. Anthropic doubled Claude Code's five-hour limits in May and bumped them again when Opus 5.5 launched in September. The weekly cap didn't grow with those changes though, so a bigger five-hour window simply lets me empty the week faster. Unfortunately for me, Opus 5.5 is my favorite model right now, and I've gotten pretty lazy about it - I even found myself using it for a spell-check the other day which is just so unnecessary. Dropping to a lighter Claude model helps a little, but it all comes out of one account's weekly cap either way, so you're still moving around inside the same bucket. Building a roster instead of picking a favorite Different pools and different jobs I never meant to pay for ChatGPT on top of Claude, but it happens to be excellent at handling some of my design work and experiments, so I've been holding onto the subscription. And that actually turned out to be a good thing for OpenCode because its connect menu has a ChatGPT option under OpenAI, and the browser login just signs you right in without having to go the API route. Usage then comes out of my plan's Codex allowance. Once that was connected, I could pick GPT-6.1 Sol from OpenCode's model selection. Anything that needs to get built now goes to GPT, or pretty much any project that starts from nothing, including notes and document work that I need to start from scratch. The rest of my model options are the free ones, and I switch between models in the same sessions with the /model command. Nemotron 3.5 Super VL picks up small edits on anything that isn't a personal project, for example, it added a search box to a little DVD tracker of mine in under a minute. Big Pickle gets the throwaway questions, where a weird answer costs me nothing. GLM-5.3-Flash ended up being the surprise. I hand it whatever GPT made and ask it to list problems without changing anything, and it's pickier than me, actually, it spotted a counter showing the wrong number whenever a search was active. Its provider runs zero retention and doesn't train on what I send, which makes me more okay pointing it at notes as well. So this spread of tasks is actually pretty simple. The premium model only touches jobs that earn it now, and all the tiny requests that used to chip away at my Opus week go to the free models. Having a different model to check the work feels more honest too. There's always a catch Free models don't come with premium polish Free models will usually be less predictable than anything you pay for. For example, Big Pickle sat on Thinking during one tiny edit until I interrupted it and handed the same prompt to Nemotron, which finished in under a minute. Claude Code almost never leaves me waiting like that, so it took a bit to get used to switching instead of letting it do its thing. There's also the privacy issue. Big Pickle's data may be used to improve the model during its free period, and NVIDIA's Nemotron endpoints are trial-only with a warning not to send anything personal or confidential, so nothing private from my vault goes to them. Also, Claude isn't on the roster at all, which I would change if I could. You can pick an Anthropic model via API, but using your Claude subscription through OpenCode isn't an option like it is for ChatGPT. Not every task deserves the best model Opus is still the model I'd keep if I could only have one, and Claude is very much still in my task bar. It just doesn't get handed every little job anymore, so the weekly meter barely comes up. But even if rate limits weren't a thing, I'd still split my work by task across different models, because that second opinion is more valuable than I thought it would be. OpenCode