OpenCode Just Took #1 From Cursor. Here's What the Adoption Crown Actually Measures.

Nine days ago I published my scorecard of the AI coding agents I actually run, and the whole premise was that "which is best" is the wrong question when 70% of us run two to four agents at once. LogRocket went ahead and answered it anyway. Their June power rankings put OpenCode at #1 — above Cursor, above Claude Code — in what they call the first major disruption to the category since Cursor 3's rebuild.

Since I graded these exact tools a week and a half ago, I owe the follow-up: what the ranking actually rewards, whether my scorecard survives it, and what — if anything — changes on my machine.

The rankings, straight

LogRocket AI dev tool power rankings — June 2026

1. OpenCode      NEW  — debuts at the top
2. Cursor        ↓    — down from #1
3. Claude Code   =    — holds position
4. Windsurf      ↓    — down from #2
5. Antigravity   ↓    — down from #4

OpenCode's case is scale. LogRocket cites 160K+ GitHub stars — the live repo reads closer to 178k as I write this — and 7.5 million developers using it every month, under an MIT license, as "the most-adopted open-source coding agent ever built." Among open-source agents it isn't even close: Gemini CLI sits around 105k stars, Codex near 90k, Cline at 63k, Goose at 48k, Aider at 46k. The crowd picked a winner while the rest of us were arguing about benchmarks.

Why the open one won

Model-agnostic is the point. OpenCode connects to 75+ providers and is loyal to none of them. In a year when open-weight models got good enough to self-host seriously and a frontier model vanished overnight by government directive, an agent harness that treats every model as swappable stops being a preference and starts being risk management. That's infrastructure, not a subscription.

The LSP loop is a real edge. LogRocket singles out OpenCode's language-server integration — compiler diagnostics fed back to the model automatically — as "unique — no other tool does this." As someone who has spent a meaningful fraction of his life pasting error output back into a model by hand, I understand exactly why that feature converts tinkerers into dailies.

It runs where clouds can't. MIT-licensed, local-first, air-gap deployable — which makes it the only serious option for regulated industries, and the natural agent for the own-your-stack crowd I belong to. The ranking reads like a referendum on flexibility over ecosystem — and flexibility won.

What an adoption crown doesn't measure

Output quality. The same LogRocket write-up notes that in blind reviews, Claude Code's output is preferred 67% of the time against Codex's 25% — and Claude Code sits at #3. Both facts are true at once because the rankings weight adoption, momentum, and flexibility, not diff quality. The most-installed tool and the tool that writes the best code are different questions, and the rankings answer the first one.

I'd also flag, honestly, which number to trust. GitHub stars measure hype velocity as much as usage — I've starred plenty of repos I never installed. The 7.5 million monthly active developers is the sturdier stat, and it's the one that should worry the proprietary tools, because every one of those developers is now used to bringing their own model.

My scorecard verdict survives contact with the rankings, in other words. Claude Code is still the quality leader I run for the work that matters; Cursor still owns the full-IDE experience (LogRocket kept it at #2 for exactly that); OpenCode is still the harness you pick when control and choice are the requirement. What LogRocket measured is that the third group is much, much bigger than anyone's pricing team assumed.

What changes in my stack

On my machine: nothing dramatic, which is itself the honest answer. Claude Code keeps the deep work. What changes is my default recommendation. When someone without a Claude subscription asks "what do I install first?", the answer is now OpenCode plus a cheap model — and with Sonnet 5 at $2/$10 introductory pricing and open-weight models a config line away, "cheap model" has never been easier to fill in. No free tier is Claude Code's known constraint; OpenCode is the pressure-release valve, and 7.5 million people found it.

The thing I'm watching next: whether the proprietary tools respond on price or on quality. Pressure from an MIT-licensed #1 is the best thing that could happen to everyone's subscription bill, including mine.

The verdict

The most-adopted coding agent in the world is now open source, MIT-licensed, and loyal to no model vendor. That's the headline, and it's deserved. It still doesn't tell you which tool writes the best diff — blind reviewers keep saying Claude Code — so the scorecard rule stands: know what each is for. What changed this month is the answer to "start where?" — and for the first time since I've been keeping score, it's an open-source answer.

Sources

The rankings, the blind-review stat, and the LSP note come from LogRocket's AI dev tool power rankings; the live star and fork counts are from the OpenCode repository; the provider count and adoption claims are from opencode.ai. The scorecard this post follows up on is mine.