Claude Code vs OpenCode: Anthropic's Built-In Containment vs an Open-Source Agent That Takes Any Model
Claude Code
L'outil de codage agentique en terminal d'Anthropic
OpenCode
The open-source coding agent for terminal, desktop, IDE and web
Verdicts par tâche
OpenCode connects to 75+ providers including local models; Claude Code's supported paths centre on a Claude account, with third-party providers on some surfaces.
Claude Code's terminal, VS Code and JetBrains surfaces also take third-party providers, so the gap is smaller than the headline.
It has a sandboxed bash tool with filesystem and network isolation, a classifier-based auto mode and isolated cloud VMs with a credential proxy.
Anthropic's own security docs still advise VMs for scripts and tool calls on untrusted content, and a dev container or sandbox runtime for extra isolation.
It is free under the MIT licence and runs on whatever provider keys or logins you already have.
You still pay for model usage; Anthropic's docs average about $13 per developer per active day for Claude Code, and your provider will bill you for OpenCode.
The docs list GitHub Actions, GitLab CI, Slack, Routines and an Agent SDK.
OpenCode's docs list an SDK and MCP servers, and we have not tested either product.
The repository is MIT-licensed, so you can audit and fork the agent.
Self-hosting means you own patching: the Sep 24, 2026 advisory (GHSA-632h-h47v-g4x4) hit npm, pnpm and Bun installs of the server and was fixed in 1.18.22.
Comparaison des fonctionnalités
| Dimension | Claude Code | OpenCode |
|---|---|---|
| IA native | IA native | IA native |
| What it is | Anthropic's agentic coding tool: one engine across terminal, VS Code and JetBrains, a desktop app (macOS, Windows, Linux beta), and web and mobile | An open-source coding agent for the terminal, a desktop app, an IDE extension and a browser-based web interface that runs against a local server |
| Models and providers | Most surfaces need a Claude subscription or an Anthropic Console account; terminal, VS Code and JetBrains also support third-party providers | 75+ LLM providers through Models.dev, including local models |
| Licence and price | Included with Pro ($20/mo, or $17/mo billed annually), Max (from $100/mo), Team Standard ($25/seat, or $20 annual), Team Premium ($125/seat, or $100 annual) and Enterprise ($20/seat plus usage at API rates). Anthropic's cost docs put the average near $13 per developer per active day | Free and MIT-licensed; you pay your model provider. The opencode.ai home page describes Zen as a handpicked set of models OpenCode has tested for coding agents and links a Go page, but gives no prices, and the Zen and Go docs pages did not load on 2026-10-12, so we do not price them |
| Permissions and sandboxing | Auto mode (a classifier model reviews actions) is the starting mode for interactive terminal and VS Code sessions; manual mode starts read-only. A sandboxed bash tool adds filesystem and network isolation, and cloud sessions run in isolated Anthropic VMs with a credential proxy | The permissions docs say most permissions default to allow; doom_loop and external_directory default to ask and .env reads are denied by default. The pages we read describe no sandbox |
| Security advisories | The first page of the anthropics/claude-code security advisories list (10 entries; GitHub shows 4 pages in all) held 6 High, 3 Moderate and 1 Low when we read it on 2026-10-12. The newest are GHSA-pq7j-f95f-qfcv (High, Oct 9, 2026, plugin commit SHA pin bypass allowing installation of attacker-controlled code) and GHSA-5j29-h97v-84ch (High, Oct 5, 2026, arbitrary file write via write-time symlink following). GHSA-7835-87q9-rgvv (High, Jun 25, 2026) is a sandbox escape via git worktree path confusion. A few entries concern Claude Desktop or CoworkVMService rather than the CLI. Fixed versions were not shown on the page, so we did not read them | The one advisory we read: GHSA-632h-h47v-g4x4 (High, CVSS 7.5, no CVE, published 2026-09-24): a cross-origin /global/upgrade request could install an attacker package; affects roughly 1.14.30 through 1.18.16 on npm, pnpm and Bun installs, patched in 1.18.22; curl, Homebrew, Chocolatey and Scoop installs are described as unaffected |
| Extensibility and automation | MCP, CLAUDE.md and AGENTS.md, skills, hooks, subagents, an Agent SDK, Routines, GitHub Actions and GitLab CI, Slack, Remote Control and Channels | Plan mode, undo and redo, /share links, AGENTS.md init, MCP, agent skills, custom tools and an SDK |
| AI-native verdict | native: the agent loop is the model, holding full context while it reads, acts and continues | native: the product is the agent loop, which plans, edits files and runs commands against any configured LLM |
Same job across terminal, desktop, IDE and browser; Claude Code adds Anthropic-hosted cloud sessions and the Claude mobile app, and one is vendor-built while the other is open source
OpenCode for freedom to pick a provider; Claude Code is the integrated path for Claude
OpenCode removes the agent subscription, not the model bill
Claude Code ships more containment, and Anthropic still points you to VMs or a dev container for untrusted work; the OpenCode pages we read describe permission rules and no sandbox, so isolate it yourself
Claude Code's first page of advisories is the longer list we read; we read neither vendor's full list, so we rank neither on security record
Claude Code has the broader automation and CI surface; OpenCode covers the core agent features
Both native
Toujours indécis entre les deux ?
Choisir est la partie facile. Le travail, c’est de le faire tourner dans votre entreprise, sur vos données, avec votre équipe qui s’en sert. Nous faisons les deux.
La conversation tourne sur Gnosari, l’un des outils de ce répertoire. Une vraie conversation, pas un script de vente.