opencode_
An open-source extension that plugs 30+ frontier AI models into GitHub Copilot Chat. You keep the native Copilot UI, tool-calling, and Agent Mode. You just get way more models, often cheaper than Copilot Pro+.
The bill that started it
On April 27, 2026, GitHub announced Copilot was moving to usage-based billing. Agentic mode, which calls the model dozens of times per task, made the cost spike unpredictable. I was paying more for fewer conversations. The premium models I relied on, Claude Opus, GPT-5.5, Gemini, were locked behind a $39/mo Pro+ tier, and the free tier was rate-limited to the point of being unusable mid-sprint.
The problem was cost, not the IDE
I liked Copilot Chat. The inline tool-calling, the Agent Mode, the context awareness. Switching to another IDE meant rebuilding my entire workflow. What I wanted was simple: keep the Copilot Chat interface, but plug in cheaper, equally capable models. DeepSeek V4, Kimi K2.6, GLM-5.1, Qwen3.7. Frontier open models that cost a fraction of what I was paying. No separate app, no context switching, no learning curve.
How it works
The extension plugs OpenCode's model gateway directly into Copilot Chat's model picker. Three tiers: OpenCode Zen gives you 2-5 rotating free models (Big Pickle is always free; DeepSeek V4 Flash, MiMo-V2.5, and others rotate weekly) plus pay-as-you-go access to Claude Opus, GPT-5.5, and Gemini. OpenCode Go ($10/mo, $5 first month) adds a curated set of open models with generous limits. Or bring your own API keys and pay nothing to OpenCode. You pick the model from the same dropdown. Everything else, the tool-calling, the Agent Mode, the inline suggestions, stays exactly where it was.
What the data says
Released May 14, 2026. In less than two months, 5,734 developers across 12+ countries installed it. Brazil, Spain, Germany, India, Japan, the US. The rating held at 5.0. The most common feedback: "I was about to cancel Copilot Pro+." The second most common: a request for one more model. Not a request for a feature. The extension already did what it needed to do.
Two extensions, one pattern
I shipped a second extension, z.ai Copilot Chat, to test whether the approach generalized. The numbers tell the story.
Same problem, same solution shape, different audience. OpenCode hit a nerve because Copilot pricing was already in the news. z.ai proved the pattern works for any model gateway, not just one. Data via VS Code Marketplace publisher dashboard, July 2026.
Why a CPO ships an open-source dev tool
Because the instinct doesn't turn off. I saw a pricing change that hurt me, assumed it was hurting others, built the smallest thing that tested that assumption, and shipped it. The extension is not a pivot. It's the same pattern: name the problem in a sentence, then build the smallest thing that proves or kills it.
What developers are saying
"Excellent extension! Works well right out of the gate and has everything I wanted it to have."
"Amazing, why I didn't find this sooner."
"Just installed the extension and loaded my OpenCode GO account and it worked. I have access to all the models in the plan."
The cheapest way to learn if a problem is real is to ship the smallest thing that solves it.