blog

Every reachpad now includes two models you can use through OpenCode without bringing an API key: Gemma 4 31B and Gemini 3.6 Flash. Model usage is included with the pad.

What is included

OpenCode comes preconfigured with both models. Open it, choose the reachpad provider, and send a prompt. There is no Google account to connect and no model bill to set up.

• Gemma 4 31B is the default reachpad model in OpenCode.
• Gemini 3.6 Flash is available from the same model picker and uses the same included allowance.

Claude Code and Codex are still installed and kept current. Those tools use your own accounts, as they did before. The two included models are another option, not a replacement for an agent setup that already works for you.

Why the pad includes models

A development environment is not ready for an agent if the first task is provisioning a provider account and putting its key into a shell. That setup is reasonable when you have settled on a provider. It is a poor first five minutes when you want to hand an unfamiliar repository to a model and see what it does.

The included models make the first prompt part of the pad. Clone a repository and open OpenCode. Ask it to explain or change the code. If one model gets stuck, switch to the other without changing credentials or moving the working tree.

How the provider key stays off the pad

We do not copy a shared upstream key onto customer machines. Each pad receives a revocable credential scoped to its owner. OpenCode presents that credential to the reachpad model gateway, which checks the allowance and adds the upstream key only while relaying the request.

The gateway meters bytes at the pipe instead of parsing prompts or responses. That gives us a daily ceiling for a runaway agent loop without adding application code that reads the conversation. The upstream key remains at the gateway, and a credential can be revoked without rebuilding the pad.

What free means

There is no per-token charge for these two models. Usage is part of every pad and does not require a separate model subscription. A daily allowance resets at midnight UTC, so one broken loop cannot consume an unbounded amount of inference. OpenCode reports a quota response if the pad reaches that ceiling.

Try it on a pad

The two models are available on new and existing pads. See the pad and workspace trial details, or create an account to start.