OpenRouter free models: use them in OpenCode, with limits

OpenRouter always has a handful of models you can call for nothing. On October 20, 2026 its public model catalog listed 468 models, 20 of them priced 0 for input and output, and 17 of those support tool calling. Tool calling is the part that matters for coding: an agent like OpenCode reads files, runs commands and edits code through tools, so a free model without tools is not much use there.

VibeiDEAll your agents. One Grid.Battery friendly, no 3D. 14 days free, no card.Download

This guide lists the OpenRouter free models that work with an agent today, shows how to run one in OpenCode, and explains the limits and the privacy setting to check before you send real code.

The free models with tool calling

Free models on OpenRouter carry a :free suffix on their ID. These are the ones with tool calling in the catalog on October 20, 2026, newest first:

The 17th is openrouter/free, a router rather than a model. Its model page says it picks a free model at random and filters for the features your request needs, such as tool calling, with up to 200,000 tokens of context depending on the model that answers.

Free models come and go, and the newest one on the list, MiniMax M3.1 Flash, arrived one day before this check. Treat the table as a snapshot and look at the catalog before you commit a project to one model.

The limits, as of October 2026

Free models are rate limited per account. The OpenRouter limits page gives two tiers:

The daily number is the one that bites with an agent. A coding task is not one request: the agent asks the model again after each tool call, so a single task can use many requests. With 50 a day you can try a model properly; for a working day of agent tasks, the 1,000 tier is the realistic one. It needs a one-time purchase of 10 credits, which then stays on your account.

Run a free OpenRouter model in OpenCode

OpenCode has OpenRouter built in. The steps from the OpenCode providers docs:

1. Create an API key in your OpenRouter account settings.
2. Start OpenCode, run /connect, search for OpenRouter and paste the key.
3. Run /models and pick a model. Many OpenRouter models are preloaded.

The model ID in OpenCode is the provider, openrouter, followed by the OpenRouter slug:

# one session on a free model
opencode -m openrouter/minimax/minimax-m3.1-flash:free

# a single task from a script
opencode run -m openrouter/minimax/minimax-m3.1-flash:free "Explain what src/queue.ts does"

A model added to OpenRouter yesterday may not be in OpenCode's preloaded list yet. First try opencode models --refresh, then opencode models openrouter to see what OpenCode knows. If it is still missing, the providers docs say you can add it as an empty entry in opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "openrouter/minimax/minimax-m3.1-flash:free",
  "provider": {
    "openrouter": {
      "models": {
        "minimax/minimax-m3.1-flash:free": {}
      }
    }
  }
}

Our guide to OpenCode models covers which setting wins when the flag, the project config and your global config disagree.

Check the privacy setting first

Free does not tell you where your prompt goes. OpenRouter's privacy and logging docs say each provider has its own data policy, and that there are separate account settings for paid and free models that control whether requests may route to providers that may train on your data. OpenRouter itself does not store prompts or responses unless you opt in to prompt logging.

Before you point a free model at client code, open your OpenRouter privacy settings and check the free-model setting. For side projects and open source code the trade is usually fine; for anything under an NDA, use a provider whose terms you have read.

OpenRouter or OpenCode Zen?

OpenCode also has its own free models through OpenCode Zen, with a data note for each one. The overlap changes quickly: MiniMax M3.1 Flash was free on both at launch, but by October 23, 2026 it had left the Zen free list and stayed free on OpenRouter. A quick way to choose:

• OpenRouter when you want the widest choice and one key for many vendors, and you can live with the daily request cap.
• OpenCode Zen when you want the models the OpenCode team lists and per-model data notes in one place. See OpenCode free models for the current list.

Where VibeiDE fits

VibeiDE is a desktop app that coordinates the Codex, Claude Code and OpenCode CLIs you already use, with your own provider accounts. You can install OpenCode from Profile > AI tool setup without needing Node.js or npm, and the info button beside OpenCode in the composer lists the current free OpenRouter models and can prefill a setup task for Claude Code or Codex (release notes). You still connect your own OpenRouter key in OpenCode. Each project has its own task queue, so small tasks for a free model wait their turn instead of all hitting the per-minute limit at once, and finished tasks wait in Ready for review. More on the setup in free AI coding models and VibeiDE for OpenCode.

Takeaways

• On October 20, 2026, OpenRouter listed 20 free models; 17 support tool calling, which an agent needs.
• Free models allow 20 requests a minute, and 50 a day until you have bought 10 credits, then 1,000 a day.
• In OpenCode, connect OpenRouter with /connect and use openrouter/<vendor>/<model>:free.
• If a new model is missing, run opencode models --refresh or add it under provider.openrouter.models.
• Check the free-model privacy setting on OpenRouter before you send private code.