Updated on
OpenCode: reviews and analysis
The open source terminal agent that works with any model provider.
Affiliate link · no extra cost to you
Our verdict
OpenCode is the open answer to Claude Code: a terminal agent with a polished interface that works with whichever provider you choose, from the subscriptions you already pay to API keys or local models. That neutrality is its argument: the tool does not decide your bill. Community development moves fast, and like any young project it has edges the commercial tools already sanded.
Best for: Anyone wanting a terminal agent without marrying any model provider.
What changed this week
Sweep of
No substantive change from the previous sweep: what we checked this week confirms what was already there.
- no change
- no change
We have tracked it since July 25, 2026: 3 sweeps on record. We publish the latest change here; the full series is not published.
What the internet says
It is still the most actively developed open terminal agent in the catalogue, shipping releases almost daily and free to run with any provider. The paid plan's access failure report that has been open since 22 July remains unresolved, and the daily flow of new issues shows no sign of slowing.
What the web repeats in favour
- Open source under MIT with development in the open, shipping a commit and a release on the very day of the sweep
- It sends far less context per turn than other agents, which is what makes real usage cheaper when you pay per token
- You can still skip any subscription and plug in your own provider, since both official plan pages repeat that subscribing is optional
- There is a cheap in-house subscription, $5 for the first month and then $10 a month, with pre-tested open models and no obligation to use it
What the web repeats against
- The paid plan piles up access failures: some subscribers see the paid models erroring while the free ones work, with the report open since 22 July and still unresolved as of today
- The pace of change is high and the queue of open issues is huge, so what works today may move next week
- The zero-retention promise has an exception declared by the vendor itself, and it lands on the free trial models in the pay-as-you-go catalogue
- It stopped bundling the plugins that reused a Claude Pro or Max subscription, which Anthropic prohibits, though Anthropic models are still available by paying per API
Sweep sources: Official pricing · Official pricing · Docs · GitHub · GitHub · GitHub
Pros / Cons
Pros
- Neutral: works with any provider or model
- Very polished terminal interface
- Free and open source
Cons
- Young project with edges to sand
- Support depends on the community
TLDR: OpenCode is the open answer to Claude Code: a terminal agent with a polished interface that works with whichever provider you choose, from the subscriptions you already pay to API keys or local models. That neutrality is its argument: the tool does not decide your bill. Free and open source, with the pace (and edges) of a young community project.
What OpenCode is and how it works
OpenCode started from a premise of independence: the terminal agent had become the craft’s central tool, and it sat in the hands of the same providers selling the models. Its offer is separating the two: an open agent, of commercial quality, working with whoever’s model you want.
The neutrality is practical, not rhetorical: it connects to the subscriptions you already pay, to any provider’s API keys or to local models, and switching engines does not mean switching tools. For anyone unwilling to marry an ecosystem, it is an insurance policy turned into a product.
The terminal interface is its other signature: polished to the detail, with an experience that does not envy the commercial products and has been an adoption hook by itself.
What it is like day to day
The flow is the genre’s standard: you describe, the agent plans and executes, you supervise and review. With a frontier model behind it, the experience gets very close to that model’s official agent, with the difference playing out in the integration details each provider reserves for its own tool.
Provider flexibility becomes habit fast: the expensive model for the hard task, the cheap one for the mechanical, the local one for the sensitive, all from the same agent without touching the flow. It is the usage pattern its users cite most.
The edges are those of a fast-growing community project: some roughness the commercial tools already sanded, documentation chasing the product and dependence on the community’s pace. Nothing that stops the profile choosing it, which tends to value exactly that openness.
Pricing and plans
Free and open source, no plans. The cost is whichever engine you connect: your existing subscription, API consumption or zero with local models. That structure makes it the category’s most flexible entry: you test with what you already have and decide later where the money goes.
Who it is for (and who it is not for)
OpenCode is for anyone wanting a first-rate agent without marrying a provider: developers with subscriptions or keys already paid, teams whose data policies demand engine control, and open source partisans for the craft’s central tool.
It is not for anyone wanting maximum polish with zero configuration, nor for anyone living happily inside one ecosystem who sees no value in neutrality.
OpenCode alternatives
Claude Code and Codex are the official agents with the deepest integration of their models. Aider is the open veteran with a more manual philosophy. Goose plays openness with broader automation ambitions. They are in OpenCode alternatives and the ranking in the best AI coding tools.
Frequently Asked Questions
Can I use my Claude or ChatGPT subscription?
It is one of its central uses: connecting the subscriptions you already pay and using their models from a neutral agent. Each connection’s details live in its documentation, and the terms are each provider’s to set.
What do I gain over my provider’s official agent?
Independence: switching models without switching tools, mixing providers per task and depending on nobody’s product decisions. If that adds nothing for you, the official agent will give you finer integration.
Is it mature enough for serious work?
Yes, with young-project caveats: the core experience is solid and the community moves fast, with some edges and documentation in pursuit. Thousands of developers use it daily.
Does it work with local models?
Yes: connected to a model on your machine, the agent runs with no per-token cost and no data leaving home. Capacity depends on the model you can run, and for bounded tasks it is already practical ground.