OpenCode Crosses 180K GitHub Stars as an Open-Source Claude Code Rival: Better UI, But Claude Code Still Wins on Cost for Subscribers
TL;DR · What you'll learn
- 1 The open-source coding agent OpenCode has crossed 180,000 stars on GitHub, positioning itself as a direct Claude Code replacement with access to hundreds of models, including free and local ones.
- 2 It ships with the same rich toolset that made Claude Code useful -- bash, edit, grep, web fetch, and web search.
- 3 Custom agents can be configured flexibly via JSON or Markdown files, illustrated with a dedicated 'code review' agent that itself runs on Claude Sonnet 4.
- 4 The verdict: the UI clearly beats Claude Code's, and on cost, OpenCode wins decisively for API-only billing, while Claude Code stays cheaper for anyone on an Anthropic subscription, since third-party harnesses are blocked there.
Read as slides
3 slides total
An Open-Source Challenger to Claude Code
OpenCode, a coding agent that lives in your terminal, has just crossed 180,000 stars on GitHub. It's positioned as a direct replacement for tools like Claude Code, 'without the glitchy UI and expensive subscriptions.' Access to hundreds of models -- including free and local ones -- means you're not locked into a single vendor's platform. It ships with the same rich toolset that made Claude Code so useful in the first place: bash, edit, grep, web fetch, and web search -- and having all of that in a free, open-source tool is, in the reviewer's view, genuinely impressive.
Flexible Custom Agent Design
One of OpenCode's strengths is how freely you can configure custom agents. If you want everything in one place, a JSON config file lets you define agents beyond the defaults -- like build and plan -- including a custom 'code review' agent that happens to run on Claude Sonnet 4. If you'd rather not keep everything in one file, you can instead describe each agent as a Markdown file inside a `.opencode` folder -- for example, defining a UI-engineer sub-agent that runs on a different model tuned for design work. An `agents.md` file, functionally equivalent to Claude Code's CLAUDE.md, teaches the agents your project's conventions.
Better UI, But Cost Depends on Your Setup
The verdict: the UI is considerably better than Claude Code's, and on top of the same toolset, you get the added benefit of bringing your own models and configuring custom tools and agents. Cost, though, isn't black and white. If you're on API-only pricing, OpenCode wins easily thanks to its access to a far wider range of cheaper models. But if you're on an Anthropic Pro or Max subscription, you can't use third-party harnesses like OpenCode at all -- Anthropic specifically blocks that access. And since that subscription bundles in a steep discount versus raw API rates, Claude Code ends up the cheaper option on pure cost for anyone already paying for it.
Editor's Take
What this comparison really exposes is that the right coding-agent choice hinges less on raw capability and more on billing arrangement. The fact that Anthropic explicitly blocks third-party harnesses for subscription users is a clear signal of ecosystem lock-in strategy at work -- and it means users now need to weigh contractual constraints alongside model performance when choosing a tool. OpenCode's 180,000 GitHub stars can be read, in part, as a measure of growing wariness toward exactly that kind of vendor lock-in.
Source
Better Stack
Open Source Is Back. Goodbye Claude.
This article auto-summarizes the YouTube video's transcript with Claude. Please refer to the original video for nuance and exact wording.
Watch on YouTube →Related
3 articles
セキグチガクの未来解剖ch
How Noda Is Driving LINE Automation to Zero Cost With Claude Code
Kazushi Noda's "Harness for zero cost": he rebuilt a 50,000-yen-per-month LINE automation tool with Claude Code and now distributes it with zero subscription and zero operating cost.
Better Stack
Claude Code Plugin "Ponytail": The "Lazy Senior" That Cuts Code by 94%
"Ponytail" is a library/plugin for Claude Code named after the archetypal veteran developer with a long ponytail and round glasses, the kind of 10x engineer who looks at 50 lines of code and silently replaces them with a single line.
Nate Herk | AI Automation
GPT 5.6 Sol vs. Fable 5, Head to Head: Fable Edges Out a Bike Game Build at 3x the Cost
With GPT 5.6's Sol, Terra, and Luna released, benchmarks show Sol beating Opus and Fable while costing far less -- but the creator decided to run both models side by side on real work for a full day instead of just quoting numbers.