OpenCode Review: The Open-Source AI Coding Agent That Took #1 on Hacker News

When an open-source project hits 1,099 points and 546 comments on Hacker News in a single day, it is not a fluke. It is a signal that the developer community has been waiting for exactly this.

OpenCode — the open-source AI coding agent built by Anomaly — crossed that threshold on March 20, 2026, becoming the #1 post on Hacker News. With 120,000 GitHub stars, 800 contributors, and over 5 million monthly developers, OpenCode has quietly become the most popular open-source alternative to Claude Code and Codex CLI. And the HN thread reveals exactly why developers are so passionate — and so conflicted — about it.

What OpenCode Is — And What Problem It Solves

OpenCode is a terminal-based AI coding agent that works with any LLM provider. Unlike Claude Code (locked to Anthropic) or Codex CLI (locked to OpenAI), OpenCode is provider-agnostic by design. You can connect Claude, GPT, Gemini, local models, or even use OpenCode’s own curated “Zen” model tier — all from the same interface.

The project is built by Anomaly, the team behind terminal.shop, and includes several people from the Charm ecosystem — the creators of Bubble Tea, the Go TUI framework that powers many modern terminal apps. This is not a weekend project. It is infrastructure built by people who care deeply about the terminal experience.

Core capabilities:

• Multi-provider support — Claude 4 Opus, GPT-5.4, Gemini 2.5 Pro, and 75+ models through Models.dev, including fully local models
• LSP integration — Automatic language server loading for intelligent code assistance
• Multi-session — Run multiple agents in parallel on the same project
• Session sharing — Generate shareable links to any coding session
• GitHub Copilot/ChatGPT login — Use existing subscriptions directly
• Build and Plan modes — Toggle between full execution and read-only analysis with Tab
• Client/server architecture — Run OpenCode on one machine, drive it from another (including a mobile app)
• Desktop app and IDE extension — Not just a terminal tool anymore, though the TUI remains the flagship

If you have been following the AI coding agent space, this feature list reads like a wish list that developers have been assembling for months.

Why Hacker News Went Wild

The HN thread is a goldmine — not because everyone loved OpenCode, but because it triggered the exact debates that define the AI coding tool landscape in 2026. Three themes dominated.

1. The Open-Source Freedom Argument

The most upvoted sentiments were variations of: finally, a real open-source alternative. Developers are increasingly uncomfortable with the vendor lock-in of Claude Code and Codex CLI. When your entire coding workflow depends on a proprietary tool tied to a single model provider, you are one pricing change away from rebuilding your muscle memory.

OpenCode lets you switch models without switching tools. That is genuinely novel in a market where every major player is trying to lock you into their ecosystem.

2. The Performance and Quality Debate

The thread became a battleground between Claude Code, Codex CLI, and OpenCode users arguing about resource usage, code quality, and TUI polish.

One commenter noted that OpenCode “often uses 1GB of RAM or more. For a TUI.” Another described the project’s development practices as “suboptimal at best,” citing a release cadence so fast that features break between versions.

But defenders pushed back: the tool works, it is free, and the pace of development — while chaotic — means problems get fixed quickly.

3. The Privacy Controversy

Perhaps the most alarming finding from the thread: OpenCode was sending prompts to Grok’s free tier by default for session title generation, even when users configured only local models. One developer ran it through mitmproxy and discovered the hidden API call, noting that Grok’s free tier trains on submitted data.

This is the double-edged sword of open source: the problem was found because the code was auditable, but it existed because the velocity-first development culture prioritized features over careful defaults.

Installation and First Impressions

Getting started is straightforward:

# One-liner install
curl -fsSL https://opencode.ai/install | bash

# Or via package managers
npm i -g opencode-ai@latest
brew install anomalyco/tap/opencode

On first launch, OpenCode prompts you to connect a provider. The /connect command walks you through authentication for OpenCode Zen, OpenAI, Anthropic, Google, or any custom provider. The /init command analyzes your project and generates an AGENTS.md file — a nice touch that helps the agent understand your codebase conventions.

The TUI itself is ambitious. Built with Bubble Tea, it offers a vim-like editor, file change tracking, session management, and a split between Build and Plan modes. Plan mode is particularly useful — it disables file modifications and lets the agent analyze and suggest without touching anything.

What impressed us: The multi-session capability is genuinely useful. Running a planning agent and a build agent simultaneously on the same project is a workflow that Claude Code and Codex CLI do not natively support.

What did not: The interface can feel overloaded. There are enough keybinds, commands, and modes that the learning curve is steeper than Claude Code’s more minimalist approach. As one HN commenter put it, it is “so full of features that I don’t really need” that it “becomes hard to use and remember how everything is supposed to work.”

OpenCode running with Gemini 2.5 Pro — showcasing the TUI and multi-model workflow.

Head-to-Head: OpenCode vs Claude Code vs Codex CLI vs GSD 2

How does OpenCode stack up against the three other major coding agent CLIs? We compared them across the dimensions that actually matter for daily use. For our detailed breakdown of the commercial options, see our GSD 2 vs Claude Code vs Codex CLI comparison.

Where OpenCode Wins

Provider freedom. This is the killer feature. No other coding agent CLI lets you switch between Claude, GPT, Gemini, and local models from the same interface. As models improve and pricing changes, this flexibility is increasingly valuable.

Cost. OpenCode itself is free. You pay only for the API calls you make — or nothing at all if you use free-tier models or local inference. Claude Code and Codex CLI charge for the tool and the API.

Customizability. Custom commands with named arguments, configurable shell environments, MCP server integration, and LSP configuration — OpenCode exposes more knobs than any competitor.

Client/server architecture. Run the agent on a powerful machine and drive it from your laptop or phone. Neither Claude Code nor GSD 2 offers this natively. Codex CLI has a similar architecture but is locked to OpenAI’s ecosystem.

Where OpenCode Trails

Model quality ceiling. When you run OpenCode with Claude Sonnet 4, you get the same model as Claude Code — but without Anthropic’s agent-specific optimizations, extended thinking integration, and the deep prompt engineering that makes Claude Code feel like it understands your project. The tool is provider-agnostic, but the best agentic experience still comes from tools tightly integrated with their model.

Stability. The fast release cadence means breakage is common. The HN thread is full of users who love the tool but wish the team would slow down and test. Claude Code and Codex CLI, backed by well-funded companies with QA processes, offer more predictable day-to-day experiences.

Ecosystem and integrations. Claude Code’s remote tasks, MCP ecosystem, and deep git integration are more mature. Codex CLI’s Rust client/server protocol shares code between its CLI, VS Code extension, and Xcode integration. OpenCode has more features on paper, but fewer of them feel polished.

Resource efficiency. Multiple HN commenters reported 1GB+ RAM usage. For a terminal tool, that is hard to justify — especially when Codex CLI achieves similar functionality at 80MB.

For a deeper analysis of how agent frameworks compare beyond CLIs, see our guide to the best open-source AI agent frameworks.

The broader AI developer tools landscape — context for where OpenCode fits in the ecosystem.

Who Should Use OpenCode — And When

Use OpenCode if:

• You want model flexibility — testing different providers on the same codebase without switching tools
• You are cost-sensitive — free tool + API-only pricing is hard to beat
• You work in privacy-sensitive environments where local-only inference matters (just audit the defaults carefully)
• You are a Neovim/terminal power user who wants maximum customizability
• You need multi-session workflows — planning and building in parallel

Stick with Claude Code if:

• Code quality is your top priority and you want the deepest model integration
• You need remote tasks, cloud scheduling, and the emerging Anthropic agent platform
• Stability and polish matter more than customizability

Choose Codex CLI if:

• Performance and resource efficiency are paramount
• You are already in the OpenAI ecosystem (GPT-5.4, ChatGPT Pro)
• You want the Rust-based client/server architecture for remote development

Consider GSD 2 if:

• You need automated multi-milestone project orchestration
• You want meta-prompting — an agent that manages other agents

The Bigger Picture: Why OpenCode Matters

OpenCode’s 120K stars are not just about the tool. They are a referendum on the state of AI developer tooling.

Developers want options. They want to use the best model for each task without being locked into a vendor’s ecosystem. They want to see the source code and audit what their tools are doing with their prompts. They want a terminal experience that respects their workflows rather than replacing them.

The HN response — 1,099 points in a day — says that the demand for an open, provider-agnostic coding agent is enormous. Whether OpenCode specifically becomes the long-term winner depends on whether the team can mature its development practices without losing the velocity that got it here.

The open-source AI agent ecosystem is entering its competitive phase. And based on what we have seen, that is exactly where it needs to be.

Have you tried OpenCode? How does it compare to your current coding agent setup? Share your experience on X @AgentConn.