VERDICT HISTORY
- Shipped OpenCode flagship card — OSS agent, Go/Zen pricing, Plan/Build modes, and Zen privacy exceptions documented.
OpenCode95FlagshipBenchmark pendingOpen-source terminal coding agent and model-agnostic harnessVerified 2026-06-15
Dossier · Build & code
OpenCode
Open-source terminal coding agent and model-agnostic harness · last verified 2026-06-15
FlagshipEDITOR'S PICK
Build & code
OpenCode
95/100
ROLEOpen-source terminal coding agent and model-agnostic harness
Editorial fit
95
Source quality
71
Citation honesty
70
Privacy controls
56
Value for money
93
Speed
80
FREEOpen source
$10/MOGo
CUSTOMZen
EDITOR'S NOTE
Desk hands-on 2026-06: Goated — the open-source agent we trust when the story is “show me the harness.” Plan mode first, Go for cheap open models, Zen when you need benchmarked frontier routes. Disable the free models that train on your prompts.
AT A GLANCE
Open-source AI coding agent for terminal, desktop, and IDE — 75+ model providers, Plan/Build modes, LSP support, optional OpenCode Go ($10/mo) and Zen pay-as-you-go gateway.
Role: Open-source terminal coding agent and model-agnostic harnessCategory: Build & codeEditorial · hands-on
Sixteen scored dimensions for the open-source coding agent — Go, Zen, Plan/Build, and provider privacy checked 2026-06-15.
PUBLIC FACTS · vendor & repo
List prices and pay-as-you-go entry points we can cite without running our own bench. Each tile links to a source when possible.
Limits & product surface
Non-price vendor claims — multipliers, caps, and API scope. Detailed matrices live in subscription and SDK sections below.
- Primary surface
- local-appCard identity ↗
- License
- Open source agentopencode.ai ↗
- Modalities
- code, textCard identity ↗
- Evidence tier
- editorial-hands-onQuality gate ↗
HOW IT WORKS · agent loop
The public positioning for this product — the loop we score against on VerdictPal.
/init the repo
Generate AGENTS.md so Plan mode knows conventions before proposing diffs.
Plan then Build
Approve the plan in chat; switch to Build only when scope is bounded.
Review with /undo
Treat each Build pass as revertible; disable free models that train on prompts.
METRIC LAB · 16 DIMENSIONS
Click a tile for the editorial note. Color follows score: coral, yellow, mint.
Shape
Avg 91 · 80–98
95Editorial fit
Open-source terminal agent with model freedom — the default we recommend for builders who refuse lock-in.
By group
- Editorial fit95
- Model breadth96
- Plan / Build modes91
- Go plan value92
- Value for money93
- Open source trust98
- Privacy posture94
- Diff discipline82
- Zen gateway quality90
- LSP integration90
- Multi-session88
- Provider flexibility97
- Terminal · desktop · IDE92
- Share & undo86
- Competitive position94
- Setup friction80
SEARCH MODES · 4 lenses
One input box, many retrieval postures. Filter by tier.
01Free
Plan mode
Read-only planning before Build touches files
02Pro
Build mode
Implementation pass after approved plan
03Max
Zen routes
Pay-as-you-go benchmarked provider pairing
04Free
BYOK
Bring your own Anthropic/OpenAI/Google keys
BENCHMARK LEDGER
Public rows are vendor or third-party claims we logged with a date. Desk rows are reserved for VerdictPal self-run results.
VerdictPal-run benches. Completed rows stay in the table; planned checks are collapsed below so unfinished work does not masquerade as evidence.
| Measure | Result | Status | Source |
|---|---|---|---|
| /init + Plan then Build on TypeScript monorepo (Go DeepSeek V4 Flash)Plan matched AGENTS.md conventions; Build diffs reviewable with /undo. | Desk hands-on | completeopencode-plan-v1 | VerdictPal desk |
Planned desk checks (1)
- Zen Sonnet route vs raw Anthropic BYOK on ten refactor promptsZen claims benchmarked provider pairing — verify latency and cost per hunk.
PRICING DECK · checked 2026-06-15
Go
$10/mo
$5 first month
- Open models bundle
- Generous limits
RAW · checked 2026-06-15: OSS agent free · BYOK any provider · OpenCode Go $5 first month then $10/mo (open models, generous limits) · Zen pay-as-you-go credits (from $20 top-up, auto-reload at $5 balance) · team workspaces in beta · card fees passed at cost (4.4% + $0.30)
RESEARCH LOG · desk notes
What we learned while building this dossier — not vendor copy.
Zen retention exceptions documented
Zero-retention on many Zen models; 30-day retention on direct OpenAI/Anthropic BYOK per docs.
COMPETITIVE LENS
Where OpenCode wins for cited research — and where a rival still belongs in the stack.
OpenCode wins when you want inspectable OSS harness, Plan/Build discipline, and Go open-model pricing; Cursor wins for tab-complete IDE speed.
OpenCode
- You want /init + Plan before any file writes
- You need Zen or BYOK routing without a closed IDE
Cursor Agent
- You live in inline completions and multi-file IDE diffs
- Your org mandates Cursor-only procurement
UNDER THE HOOD
Vendors named on the product about page — useful for procurement and privacy reviews.
- Taste terminal alt
- Command Code
- IDE agent
- Cursor
- Factory orchestration
- Factory
DEEP PANES · 6 LENSES
Editorial lenses only. Subscription and API pricing live in their own sections above.
Lens 01
Open source harness
OpenCode is the agent you can read — terminal, desktop beta, and IDE extension share one OSS core with massive community velocity.
- Signal
- 160K+ GitHub stars cited on site
- Pin a release hash for student labs
- Read AGENTS.md /init output before trusting plans
TEST SCENARIOS · hands-on lab
How we exercised the product. Step through each run before you trust the scores.
Test run
Step 1 of 3
Install opencode; subscribe to Go; /connect
BEST FOR
- Developers who want an inspectable open-source agent instead of a closed IDE-only copilot
- Teams that need Copilot, ChatGPT, Anthropic BYOK, Zen, Go, or local models in one harness
- Students and researchers on a budget using Go or free Zen models with spend caps
- Privacy-sensitive repos where vendor storage claims matter — read Zen model exceptions first
AVOID IF
- You need a polished GUI IDE with Tab-as-primary workflow — Cursor may fit better
- You will run Build mode on production branches without tests, Plan review, or /undo discipline
- You cannot track Zen auto-reload ($20 when balance hits $5) and monthly workspace caps
- You need guaranteed EU-only inference on every model without admin model allowlists
STRENGTHS
- curl / npm / brew / Docker install
- Terminal TUI, desktop app beta, IDE extension
- Plan mode (read-only) and Build mode (writes)
- /init project analysis and AGENTS.md generation
- LSP auto-load for typed context
- Multi-session parallel agents
- /share session links, /undo, /redo
- GitHub Copilot and ChatGPT account login
- 75+ providers via Models.dev and /connect
- OpenCode Zen pay-as-you-go gateway (BYOK compatible)
- OpenCode Go subscription for open models ($5 first month)
- Custom commands, themes, formatters, MCP
WEAKNESSES
- Build mode can apply diffs that pass typecheck but fail course rubrics or security review
- Zen free/promo models may retain data for training — Big Pickle, DeepSeek Flash Free, Nemotron Free, etc.
- Auto-reload and monthly caps interact confusingly when balance drops below $5
- Multi-session parallelism can stomp the same files without branch discipline
- Desktop app is still beta — terminal remains the most documented surface
- Model deprecation table moves fast — pinned configs break on schedule
EDITORIAL EVIDENCE · 3 ENTRIES
TASK
Run /init + Plan mode on a TypeScript monorepo, then Build the smallest feature with Go-plan DeepSeek V4 Flash.
Desk hands-on: Plan output matched repo conventions after AGENTS.md; Build diffs were reviewable with /undo available.
↳ Time-to-merge, test pass rate, Zen credit burn, wrong-file edits.
TASK
Compare OpenCode Zen Sonnet route vs raw Anthropic BYOK on the same ten refactor prompts.
Pending formal bench; Zen marketing claims benchmarked provider pairing.
↳ Latency, cost per accepted hunk, provider retention path.
TASK
Run two parallel sessions on one branch — measure file collision rate.
Pending.
↳ Branch discipline, session naming, merge conflict cleanup time.
PRIVACY DEEP-DIVE · checked 2026-06-15
OpenCode states the agent does not store your code or context data. Zen routes prompts to US-hosted providers with zero-retention claims on many models; exceptions include free/promo models (Big Pickle, DeepSeek Flash Free, MiMo, Nemotron Free, North Mini Code Free) and standard 30-day retention on direct OpenAI/Anthropic API paths. Taste-like project files (AGENTS.md) stay in your repo. Billing via card with fees passed at cost.
Training on your dataNot stated
EU data residencyNot documented
SOC 2 attestationNot documented
Local-first by defaultCloud only
ALTERNATIVES
Head-to-head
Compare side-by-side → open compare
WORKFLOW ROLES
How this tool fits into a composed research stack:
open-source coding agentplan-then-build refactorsmulti-provider model routing
QUALITY GATE · FLAGSHIP
Evidence ready
Benchmark pending
- desk hands-on 2026-06-15: goated open-source agent card from site docs + desk trial.
- Public recommendation: true
SOURCE LEDGER · 5 checked
- Product sitechecked 2026-06-15
- Docs introchecked 2026-06-15
- Zen pricingchecked 2026-06-15
- Go planchecked 2026-06-15
- Providerschecked 2026-06-15
Related dossiers
The Pack · Editorial newsletter
New cards in your inbox. Free.
One short email when a card ships or changes status. No tracking, no third-party analytics. Unsubscribe in one click.