| Rating |
|---|
| AgentsWisdom score | 75 | 73 |
|---|
| Confidence | high | high |
|---|
| Rank in category | No. 1 of 16 | No. 3 of 16 |
|---|
| Pillars |
|---|
| Adoption & momentum | 95 | 93 |
|---|
| Experts (independent reviews) | 71 | 73 |
|---|
| Crowd (community sample) | 53 | 46 |
|---|
| Community posts sampled | 23 | 29 |
|---|
| Experts vs crowd | Crowd cooler by 17 | Crowd cooler by 27 |
|---|
| Adoption |
|---|
| GitHub stars | 210k | 91k |
|---|
| npm weekly downloads | 2.2M | n/e |
|---|
| PyPI monthly downloads | n/e | n/e |
|---|
| VS Code installs | n/e | n/e |
|---|
| Reported users | 16M | n/e |
|---|
| Reported customer organisations | n/e | n/e |
|---|
| Releases, last 90 days | 42 | 46 |
|---|
| Benchmarks (shown, not scored) |
|---|
| Terminal-Bench 2.0 | No. 36 of 43 · 51.69%
OpenCode + Claude Opus 4.5 | not listed |
|---|
| SWE-bench | not listed | not listed |
|---|
| Independent reviews | 9 | 8 |
|---|
| Trust & safety |
|---|
| Overall grade | B | A |
|---|
| Permission model | C | A |
|---|
| Data access scope | A | A |
|---|
| Data storage | A | A |
|---|
| Data retention & training | n/e | A |
|---|
| Incident history | C | C |
|---|
| Compliance | D | B |
|---|
| Transparency | A | A |
|---|
| Price |
|---|
| Price | Free, paid from $10 | Free, paid from $10 |
|---|
| Plans | - Free (BYOK): Free
- OpenCode Go: $10 / monthly
- OpenCode Zen: Custom
- Enterprise: Custom
| - Personal: Free
- Pro: $10 / monthly
- Business: $30 / monthly
|
|---|
| Product |
|---|
| Category | Coding agents | Coding agents |
|---|
| Maker | Anomaly | Zed Industries |
|---|
| Licence | MIT | GPL-3.0-or-later |
|---|
| Use it via | CLI, Desktop app, IDE extension, Web app | Desktop app |
|---|
| Runs on | macos, linux, windows | macos, linux, windows |
|---|
| Models it drives | Claude, GPT, Gemini, Kimi, DeepSeek, GLM, Qwen, local models, any OpenAI-compatible | Claude, GPT, Gemini, DeepSeek, Mistral, Amazon Bedrock, GitHub Copilot models, any model via OpenRouter / Vercel gateway, local via Ollama / LM Studio, external ACP agents (Claude Agent, Codex, Gemini CLI, OpenCode, Copilot, Cursor, Pi, Poolside) |
|---|
| Key features | - Terminal UI, desktop app and IDE extension
- Bring-your-own-key across many providers plus local models
- Plan and Build agent modes
- LSP integration
- Configurable allow/ask/deny permissions per tool, with external_directory scoping
- Optional session sharing via /share
- Optional paid model access via OpenCode Zen and OpenCode Go
| - Agent Panel with built-in agent: file edits, terminal, fetch, MCP tools
- Parallel agent threads in a Threads Sidebar, optional per-thread Git worktree isolation
- Hosts external agents over the Agent Client Protocol, with an ACP Registry for installing them
- Per-tool permission rules (allow / confirm / deny) with pattern matching on commands and paths
- OS-level sandbox for terminal and fetch tools on by default since v1.14 (Seatbelt on macOS, Bubblewrap on Linux, WSL on Windows)
- Checkpoints to restore code, multibuffer review of agent edits, follow-the-agent view
- Agent profiles to choose which built-in and MCP tools are available
- Edit predictions (inline next-edit completions) alongside the agent
|
|---|
| Verdicts |
|---|
| Praised for | - Model freedom: one config change to switch provider or run locally (Composio, heyuan110)
- Low-cost access through the $10 Go plan (Thomas Wiegold; multiple HN users)
- Works reasonably well across many models out of the box (HN)
| - Can run and switch between several agents (built-in and ACP external agents) in one editor window
- Parallel agent threads with worktree isolation are praised by users moving from single-thread agent workflows
- Fast, lightweight editor that users keep even when they run agents elsewhere
- Free bring-your-own-key use and external agents on the $0 plan
|
|---|
| Criticised for | - Fast release cadence brings regressions (AICoderScope)
- High memory/CPU use reported (GitHub Memory Megathread; HN)
- Repeated serious security advisories in 2026, including unauthenticated command execution
- Security track record judged weaker than Claude Code (heyuan110)
| - Agent UI and parallel-agent layout described as confusing by several users after the April 2026 changes
- Built-in agent seen by some reviewers and users as less capable or less turnkey than Cursor, Claude Code or OpenCode
- ACP-hosted external agents reported as buggy or limited (buffer not refreshing, lost LSP access, never-ending runs)
- Diff/code-review tooling for reviewing agent output considered insufficient by some users
|
|---|
| Evidence |
|---|
| Evidence collected | 26 Sept 2026 | 26 Sept 2026 |
|---|
| Sources cited | 8 | 10 |
|---|