Skip to content
Research

Research

How to use OpenRouter free models to stretch your AI bot's usage

ResearchedWhat this sticker meansAdvancedWork

Last updated:

Quick answer

Keep planning on your main bot. Hand summaries, sorting, boilerplate, and first drafts to Pi on an OpenRouter free model and to OpenCode.

At a glance

All outcome figures are estimates from cited sources, not independent verification by us.

What it is allowed to do
Drafts for you

Writes drafts (emails, posts, plans, replies). You review and send. Nothing goes out without you.

Reviewed Oct 20, 2026

Time saved (estimate)
Not measured yet
Difficulty
Advanced
Tools needed
OpenRouter, Pi, OpenCode
Typical cost
OpenRouter's docs cap :free models at 20 requests a minute and 50 a day, or 1,000 a day once you've bought at least 10 credits. OpenCode marks its free models as limited-time, so the lineup changes. Other tool costs are not stated in sources.

Basis (estimate): Not measured yet. Their claim, no numbers shared.

What does this setup do?

Keep the planning and the final call on your main bot, and hand grunt work (summaries, file sorting, boilerplate, first drafts) to Pi on an OpenRouter free model and to OpenCode.

Who is it for?

People whose bot hits its usage limit right when they need it. A lot of what it does is grunt work: summarizing, sorting, first drafts. Grunt work does not need the top model.

How do you set it up?

Tools

  • Main bot: keeps the planning and the final call. It decides what to hand off.
  • Helper 1, Pi on OpenRouter: Pi is an open-source coding agent you run in a terminal. Log in with an OpenRouter key and pick a model whose ID ends in :free.
  • Helper 2, OpenCode: another terminal coding agent. Its Zen catalog lists some models as free for a limited time.
  • Radar (optional): once a week it checks which free models are new and ranks them for the jobs you hand off most.

Steps

  1. Main bot: keeps the planning and the final call. It decides what to hand off.
  2. Helper 1, Pi on OpenRouter: Pi is an open-source coding agent you run in a terminal. Log in with an OpenRouter key and pick a model whose ID ends in :free.
  3. Helper 2, OpenCode: another terminal coding agent. Its Zen catalog lists some models as free for a limited time.
  4. Radar (optional): once a week it checks which free models are new and ranks them for the jobs you hand off most.
  5. Give the starter prompt to your main bot.

Example output

Illustrative. Your results will differ.

  • Hand-off jobs go to a free model. Planning stays with the main bot.
  • A rate-limit error gets one retry, then the main bot does the task.
  • Once a week: new free models, the top 3 ranked, and a question before any switch.

Where to paste this prompt

ChatGPT: ChatGPT → Project instructions, or ChatGPT Tasks for scheduled reminders (once a day on Free)

Copy-paste prompt

Adapt bracketed fields to your situation.

You have two helpers for cheap work: Pi (using an OpenRouter :free model) and OpenCode (using one of its free models).
Before you start a task, sort it:
- Hand off: summaries, file sorting, boilerplate code, first drafts, reformatting.
- Keep: anything with passwords, keys, personal data, client data, or a final decision.
When you hand off, send only the text the helper needs. Check its result before you use it.
If a helper returns a rate-limit error (429), wait and retry once, then do the task yourself.
Once a week, list any new free models and rank the top 3 for my hand-off jobs. Ask me before switching.

Newsletter

Get the free starter pack

Free starter pack with five beginner copy-paste prompts after you confirm, plus three sourced setups in your inbox each week. Unsubscribe anytime.

What usually goes wrong?

Where agents tend to stall on this job. More patterns in Where AI gets stuck.

  • Last checked: Oct 20, 2026. The free tiers have limits. OpenRouter's docs cap :free models at 20 requests a minute and 50 a day, or 1,000 a day once you've bought at least 10 credits. OpenCode marks its free models as limited-time, so the lineup changes.

Which chat tools work with this prompt?

Paste the prompt into any of these assistants (free or paid tiers).

  • ChatGPT
  • Claude
  • Claude Code

Common questions

Does OpenRouter have free models?
Yes. Free variants have model IDs ending in :free. Which ones exist changes often, so check the models list.
What is the OpenRouter free models rate limit?
Per OpenRouter's docs: 20 requests per minute. Daily it's 50 requests, or 1,000 if you've bought at least 10 credits.
Are OpenRouter free models safe for private data?
Not by default. Some providers may log or train on prompts. Turn on the setting that blocks training providers, and keep private data off free models.
Does OpenCode have free models?
Yes, its Zen catalog lists some models as free for a limited time. They may only work inside the OpenCode app and not through the API.

What tips make it smoother?

  • Some providers behind free models may keep or train on your prompts. OpenRouter has separate privacy settings for free and paid models, and you can block providers that train.
  • Never send keys, client files, or anything personal to a free model.
  • If a helper returns a rate-limit error (429), wait and retry once, then do the task yourself.

Where do the claims come from?

Summaries use public sources linked below, not independent verification. How setups are picked.

  • Read @DegenYan9’s write-up (x.com)

    @DegenYan9: Says they connected Pi (on OpenRouter) and OpenCode (with its free models) to their bot so it hands tasks to both. They say a radar ranks new free models for Pi, and that this gets them a lot more total tokens. Their claim, no numbers shared.

Related setups

  • Debug a Python script with AI

    Debug a Python script with AI: paste the traceback and the code, get a plain-words explanation of the cause and one fix to test.

  • How to give your AI bot a long-term memory with an Obsidian vault

    Replace fuzzy in-chat “memory” with inspectable Markdown files: the bot appends a daily note and decision log to a Git-backed vault; you review in Obsidian; a weekly routine sorts an Inbox folder into PARA (Projects, Areas, Resources, Archive).

  • Use Gemini inside Google Sheets

    Use Gemini inside Google Sheets: AI formulas that categorize and summarize row by row, plus a side panel that builds sheets and dashboards from prompts.

Best AI setups for solo operators (2026) →

Ready for the next one?

Next up: a Research setup on the Personal path.

Debug a Python script with AI

Debug a Python script with AI: paste the traceback and the code, get a plain-words explanation of the cause and one fix to test.

Estimate: 30 to 90 minutes of stuck-on-an-error time per bug.