OpenCode+27 Positive

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt OpenCode?

Should your team standardize on OpenCode when its defaults let the agent act without asking?

30-Second TakePick OpenCode if your team runs several model vendors and wants one agent it can script.
Best fitPlatform-minded teams that want one agent across Anthropic, OpenAI, Google, Bedrock, open, and local models, run from terminals, editors, and their own pipelines.GILD Review
Consider another workflowIf engineers will run Claude all day on a Claude subscription, Claude Code fits better; if you want a sandbox on by default, use Codex CLI.GILD Review

Drawn from 111 sources and 8 operator insights

GILD analysis

OpenCode Review

OpenCode is one open-source agent for every model a team pays for, Claude and GPT on API keys or open models on the US$10 Go plan, with a scriptable server and SDK. The team owns the risk, since OpenCode 2 allows every action except .env reads and ships no sandbox, and OpenCode published three security advisories and a Console breach notice in 2026. Adopt it when a platform owner will pin versions and enforce a permission policy.

AI Coding Agents
Security
AI Costs

The decision

Best fit: Platform-minded teams that want one agent across Anthropic, OpenAI, Google, Bedrock, open, and local models, run from terminals, editors, and their own pipelines.
Consider another workflow: If engineers will run Claude all day on a Claude subscription, Claude Code fits better; if you want a sandbox on by default, use Codex CLI.
Adoption condition: Pin the version and enforce a deny-first permission policy through Enterprise central config. Run agents in containers with only the project mounted.

Basis: OpenCode and Anomaly documentation and security advisories checked Oct 25, 2026, read beside 90 public accounts. We ran no performance benchmark.

Buying & deployment factsWhat engineering leaders should verify before a broader rollout.

How you buy it

Free, MIT license

Models come from your own keys, a Copilot or ChatGPT sign-in, Zen, or Go at US$10 a month.

Source 3, Anomaly, Download: Download Source 6, Anomaly, Providers: Providers Source 1, Anomaly, OpenCode Go: OpenCode Go Source 2, Anomaly, Enterprise: Enterprise

Details

The agent is free under the MIT license. Models come from your own provider keys, a GitHub Copilot or ChatGPT Plus or Pro sign-in, OpenCode Zen pay-as-you-go, or OpenCode Go at US$10 per month (Go Plus US$40 per month). Enterprise is priced per seat through sales.

Cost drivers

Model usage

Go allows 20% of the monthly allowance per 5 hours; Zen adds US$1.23 per US$20 top-up.

Source 1, Anomaly, OpenCode Go: OpenCode Go Source 5, Anomaly, OpenCode Zen: OpenCode Zen Source 2, Anomaly, Enterprise: Enterprise

Details

Model usage is the cost. Go limits run on rolling windows of 20% of the monthly allowance per 5 hours and 50% per week. Zen passes through provider prices, plus a US$1.23 card processing fee on each US$20 top-up. Enterprise charges no tokens when you bring your own gateway.

Where it runs

Terminal, desktop, web

Plus VS Code, Cursor, Zed, Windsurf, and VSCodium extensions and GitHub Actions.

Source 3, Anomaly, Download: Download Source 7, Anomaly, GitHub integration: GitHub integration Source 8, Anomaly, Anomaly: OpenCode 2.0: Anomaly: OpenCode 2.0

Details

Terminal UI, desktop app for macOS, Windows, and Linux, a web client, extensions for VS Code, Cursor, Zed, Windsurf, and VSCodium, and GitHub Actions through /opencode comments. In OpenCode 2 every client attaches to one background server.

What to verify

Allow by default

OpenCode 2 allows every action except .env reads, and outside paths need no prompt.

Source 4, Anomaly, OpenCode 2 permissions: OpenCode 2 permissions Source 3, Anomaly, Download: Download Source 6, Anomaly, Providers: Providers

Details

OpenCode 2 allows every action by default and asks before .env reads, and paths outside the project proceed without a prompt. Confirm data retention for any Go or free model you allow, and that no one connects a Claude Pro or Max plan.