Rating · evidence to 26 Sept 2026
- No. 1 of 16 in category
- High confidence
- Trust B
- Open source
The Register / Coding agents / OpenCode
Register entry · Coding agents · updated 26 Sept 2026
Anomaly · Coding agents
OpenCode is an MIT-licensed coding agent that runs in the terminal, as a desktop app or via an IDE extension, and connects to many model providers or local models. Anomaly also sells optional model access: OpenCode Zen (pay-as-you-go curated models) and OpenCode Go ($10/month subscription to open-weight coding models). Most tool permissions default to allow; access outside the project directory defaults to ask.
Rating · evidence to 26 Sept 2026
| Price | Free, paid from $10 |
|---|---|
| Use it via | CLI, Desktop app, IDE extension, Web app |
| Runs on | macos, linux, windows |
| Licence | MIT |
| Updated | Evidence collected 26 Sept 2026 |
Weights for Coding agents: adoption 35%, experts 40%, crowd 25%. Pillars without enough evidence are left out and the others re-weighted. The method.
| Adoption & momentumweight 35% | 95 | GitHub stars 210k → 100 · npm weekly downloads 2.2M → 90 · Reported users 16M → 87 · 42 releases in 90 days → 100reach × 0.7 + momentum × 0.3 |
|---|---|---|
| Expertsindependent reviews · weight 40% | 71 | 9 independent reviews (5 positive · 4 mixed · 0 negative), averaged toward 50 for small samples. Benchmarks, shown not scored: Terminal-Bench 2.0 No. 36 of 43 (OpenCode + Claude Opus 4.5) |
| Crowdcommunity sample · weight 25% | 53 | 23 dated posts from hackernews, forum: 11 positive · 3 mixed · 9 negative (positive + ½ mixed) ÷ n, averaged toward 50 for small samples |
| Rating | 75 | high confidence (3 of 3 pillars, 23 community posts). |
Evidence checks. 1 GitHub issue not counted as sentiment. 1 item could not be checked at the source (the page could not be fetched, or the quote is from a comment under the linked post); kept and marked “unverified”. How evidence is checked.
Editorial adjustment. Incident-history re-grade under methodology revision 2 (see the trust section).
Seven dimensions graded A–D from documented facts. Overall: B.
Docs: most permissions default to "allow"; only doom_loop and external_directory default to "ask". No sandbox; allow/ask/deny configurable per tool. source
Access outside the project directory is gated by the external_directory permission (default ask) with path-pattern allowlists. source
Runs locally; enterprise docs state OpenCode does not store code or context data. Optional /share uploads conversations to opencode.ai. source
Depends on provider. Zen docs: most Zen providers zero-retention/no training; OpenAI and Anthropic requests retained 30 days; some free models may use data for improvement. Mixed, so not graded. source
Three GitHub security advisories in 2026 (two high, one critical), all with patched versions. Regraded from D: Revision 2: serious issues that were each fixed and publicly disclosed grade C; D is kept for unresolved issues or slow or undisclosed fixes. source
No SOC 2 / ISO 27001 attestation found in enterprise docs. source
MIT licence, public GitHub releases. source
9 named independent reviews found; dated reviews from the last 12 months are scored.
Pick OpenCode if you want a low-cost alternative that lets you use any and all models.Prathit Joshi (Composio blog) · 11 Jun 2026 · composio.dev
OpenCode is the most serious open-source terminal coding agent available right now.AICoderScope · 29 May 2026 · aicoderscope.com
Security is a top priority (Claude Code’s track record is cleaner)heyuan110.com · 13 Mar 2026 · heyuan110.com
OpenCode Go is the cheapest multi-model AI coding subscription available in 2026.Thomas Wiegold · 15 Apr 2026 · thomas-wiegold.com
OpenCode has closed the feature gap considerably. For everyday tasks — multi-file edits, shell commands, codebase search, refactors — it's hard to tell the difference from Claude Code.Jason Zhou (AI Builder Club) · 28 May 2026 · aibuilderclub.com
Why OpenCode Beat Out Every Other AI Coding Harness I TriedHamza Boulahia (Towards AI) · 6 Aug 2026 · unverified · pub.towardsai.net
OpenCode constructs an argument that I find difficult to refute: the terminal is not a compromise.Murat Akdeniz (Codexpedite) · 20 May 2026 · codexpedite.com
OpenCode is a brilliant executor, but a terrible architect.RuView Blog · 19 Jun 2026 · ruview.pro
While OpenCode + Oh-My-OpenCode has impressive performance, there's a lot of dead time where nothing is happening.Chandler (chand1012, DEV Community) · 23 Jan 2026 · dev.to
23 coded posts from hackernews, forum.
enjoying my $10 OpenCode Go sub, dual OEM spark setup, and open weights.Hacker News comment · 25 Sept 2026 · news.ycombinator.com
I may end this and go back to my own, opencode is so sloppy, but then what harness isn't?Hacker News comment · 23 Sept 2026 · news.ycombinator.com
opencode, pretty good setup out of the box, no need to go through and find packages for the basics like piHacker News comment · 23 Sept 2026 · news.ycombinator.com
I do not know why but opencode pushes CPU usage to very high which has NOT happened with DeepSeek harness even once.Hacker News comment · 22 Sept 2026 · news.ycombinator.com
OpenCode works pretty well for all modelsHacker News comment · 23 Sept 2026 · news.ycombinator.com
session management - accurate (unlike OpenCode last time I tried)Hacker News comment · 23 Sept 2026 · news.ycombinator.com
Love Pi because it does a good job managing the context, open code meh, claude code nope.Hacker News comment · 23 Sept 2026 · news.ycombinator.com
your system will probably be pwned soon since it's JS + NPM + vibe codingHacker News comment · 24 Apr 2026 · news.ycombinator.com
As published, checked 26 Sept 2026.
| Plan | Price | Notes |
|---|---|---|
| Free (BYOK) | Free | MIT-licensed agent; pay your own model provider. Some free models offered via Zen. |
| OpenCode Go | $10 / monthly | Subscription to 30+ open-weight coding models with usage limits per 5-hour window. |
| OpenCode Zen | Custom | Pay-as-you-go per 1M tokens; card fees passed at cost (4.4% + $0.30). |
| Enterprise | Custom | Per-seat; contact sales. |
Other ranked entries in Coding agents.
Every figure above traces to one of these, observed up to 26 Sept 2026.