OpenHands pricing, free plan and limits
Open-source AI agent that plans, codes, tests, and ships pull requests
- Best for
- Backend and platform engineers self-hosting a coding agent instead of depending on a single vendor's managed service
- Not for
- Non-technical operator wanting a zero-setup coding assistant
Should you use OpenHands?
OpenHands suits engineers who would rather run and audit their own coding agent than hand that control to a single managed vendor, especially anywhere proprietary code can't leave company infrastructure. It connects to six Git providers, from GitHub to Azure DevOps, and rewards teams willing to operate and maintain their own agent stack.
Model-agnostic self-hosting: it is one of the few full coding-agent platforms (not just an IDE plugin) that a regulated enterprise can run entirely inside its own VPC with its own choice of LLM.
The moat is community and switching-cost avoidance, not lock-in: 89,000+ GitHub stars, 500 contributors and MIT licensing make OpenHands the default open alternative, but the same openness means a well-funded rival (Cognition's Devin, GitHub Copilot Workspace) can copy any feature and a self-hoster owes OpenHands nothing once running.
Also for
- Security and compliance teams at regulated companies that cannot let proprietary code leave their own infrastructure
- Technical founders and small teams optimizing model cost by switching between GPT, Claude, Gemini, or open-weight models
Show 3 more
- Engineering platform teams standardizing agent governance (SSO, RBAC, spend budgets) once more than one developer runs agents
- Platform engineer standing up an internal agent platform
- Founding engineer at a seed-stage startup optimizing LLM spend
Also not for
- Enterprise buyer who wants a single accountable vendor for support and liability
Main trade-off
Output quality and reliability are only as good as whichever LLM the operator connects, so the product owns no consistent quality bar the way a single-model competitor does.
OpenHands pricing
- Top plan
- $500/mo
- Billing model
- Freemium
Agent Canvas (open source, MIT): free, runs on your laptop or a shared VM, bring your own LLM key, use any agent. Teams: $500/month for up to 10 users, +$50/month per extra seat, self-hosted, with SSO/SAML, RBAC and budgeting controls; 30-day trial; capped at 5,000 sessions per month. Enterprise: custom quote, unlimited seats and users, enterprise agreement and support tiers. OpenHands Cloud (app.all-hands.dev) is still offered, but its plan pricing is not listed on the pricing page.
Before you commit
- Cloud LLM token spend during long autonomous sessions can exceed what a casual user budgets for, especially on ambiguous tasks that cause looping.
- Self-hosting requires ongoing infrastructure and Docker/Kubernetes maintenance that a managed competitor's fee already covers.
What OpenHands does on its own
- Scope of autonomy
- Semi-autonomous
- Human approval boundary
- Human in the loop
- Maturity
- Generally available
Safety controls
- Runs sandboxedSupported
- Audit loggingSupported
- Human in the loopSupported
- Output filteringNot supported
What it can operate
- Web browsing
- Code execution
- File system
- API calls
- Shell commands
Where OpenHands runs
- Deployment
- Cloud · Local · Hybrid · API
- Where it runs
- Web · CLI · API
- Memory
- Long term
- Context window
- Model-dependent (no fixed limit; determined by whichever LLM is connected)
Models it runs on
- GPT (OpenAI)
- Claude (Anthropic)
- Gemini (Google)
- GLM (Zhipu)
- Qwen
- DeepSeek
- Local open-weight models via Ollama/vLLM
Systems it connects to
| System | Type |
|---|---|
| GitHub | Native |
| GitLab | Native |
| Bitbucket | Native |
| Azure DevOps | Native |
| Jira | Native |
| Slack | Native |
| Model Context Protocol (MCP) servers | API |
Benchmark scores
- OpenHands Index - Issue Resolution (top-scoring backend, claude-fable-5 on SDK v1.18.1, 2026-06-11)
- 95.8 / 100
- OpenHands Index - Overall Average (top-scoring backend, claude-fable-5/GLM-5.1 tie, SDK v1.18.1, 2026-06-11)
- 81 / 100
What OpenHands does
OpenHands is an open-source, model-agnostic AI coding agent that has passed 89,000 GitHub stars as of September 2026. Given a task or GitHub issue, it plans, writes, tests, and opens a pull request without step-by-step approval, running GPT, Claude, Gemini, or open-weight models locally, in the cloud, or self-hosted inside your own infrastructure.
End-to-end issue resolution
Hand it a GitHub issue and the CodeAct agent works out a fix, checks it against the existing test suite, debugs whatever breaks, and submits the result as a ready-to-review pull request unattended.
Model-agnostic by design
Runs closed models like GPT, Claude, and Gemini or open-weight models like Qwen and DeepSeek, connected through your own API key or served locally via Ollama or vLLM.
Six native Git integrations
Connects to GitHub, GitLab, Bitbucket Cloud, Bitbucket Data Center, Forgejo, and Azure DevOps to open pull requests and merge requests directly from the agent's workspace.
MCP tool support on every plan
Ships with Model Context Protocol support on the free Open Source tier and every Cloud plan, letting the agent call external tools and data sources it wasn't originally built with.
Show all 6 features
Beta Planning Mode
A dedicated Plan Mode drafts a structured PLAN.md file outlining implementation steps before switching to Code Mode to execute them, added in a March 2026 update.
Enterprise Agent Control Plane
Centralizes SAML/SSO, role-based access, and per-user or per-team spend budgets so security and platform teams can govern agents once more than one developer is running them.
Strengths and limits
Strengths
- Fully open source under the MIT license, so any team can self-host it, inspect the agent's exact behavior, and keep proprietary code off a third-party server, unlike closed platforms such as OpenAI Codex.
- Model-agnostic across more providers than most rivals: OpenAI, Anthropic, Google, Qwen, DeepSeek, and any locally served open-weight model, so teams can pick the cheapest or best-performing option for a given task instead of being locked to one vendor's model.
- Backed by an active open-source community: more than 89,000 GitHub stars and 500 contributors as of September 2026, with a fast-moving Enterprise release cadence that rarely goes two weeks without a new version.
Limits
- Bring-your-own-key use with a frontier model for long autonomous sessions can get expensive fast, on top of the plan price once a team outgrows the free tier.
- Self-hosting demands real infrastructure setup and ongoing maintenance, and output quality depends heavily on which LLM you connect, an inherent tradeoff of being model-agnostic rather than a single tuned model.
- Ambiguous or long-horizon, multi-repo tasks can cause planning drift or looping that needs human intervention; independent reviews describe it as rougher around the edges than polished commercial agents for teams wanting a zero-setup experience.
OpenHands alternatives
| Agent | Where it runs | Price | Maturity | Checked |
|---|---|---|---|---|
| OpenHands This page | Cloud · Your machine · API | Free plan | Available | 3 Nov |
 Claude CodeClaude Code wins on out-of-the-box polish with Anthropic's own models; OpenHands trades that for the ability to run any LLM and to self-host the whole stack. | -- | -- | -- | -- |
 OpenAI CodexOpenAI Codex is the easier on-ramp for a team standardized on ChatGPT; OpenHands is the better fit once you need to self-host or bring in a non-OpenAI model. | Cloud · Your machine · API | Free plan | Enterprise | 6 Nov |
 ClineCline stays inside VS Code as a lightweight extension; OpenHands is a full platform with a web GUI, CLI, and Enterprise controls for teams running agents at scale. | Your machine · API | Free plan | Available | 27 Sep |
Also compared
Gemini CLIGemini CLI is built around Google's own models; OpenHands adds the option to route the same work through GPT, Claude, Qwen, DeepSeek, or a self-hosted open-weight model instead.
OpenHands facts
- Works offline
- Yes
- Minimum skill level
- Intermediate
- Budget fit
- Budget
- Open source
- Yes
- Ships
- Weekly
- Last major update
- Maker founded
- 2024
- Headquarters
- Boston, Massachusetts
- Team size
- 25-50
- Total funding
- $23.8M total ($5M seed Sep 2024 led by Menlo Ventures; $18.8M Series A Nov 2025)
What changed
Questions about OpenHands
What does OpenHands actually cost?
OpenHands Agent Canvas is free and open source under the MIT license, and you bring your own LLM key. Teams costs $500 per month for up to 10 users, with $50 per month for each extra seat, and includes SSO/SAML, role-based access and budgeting controls. Enterprise is quote-only with unlimited seats. The pricing page, read on 2026-11-03, no longer lists a separate Cloud Individual plan.
What do you get on OpenHands's free tier?
The free Agent Canvas tier is open source and self-hosted, so you run it on your laptop or a shared VM with your own LLM key. It supports any agent, including Claude and Codex, plus automations and MCP tools. It has a single shared identity, and it leaves out the multi-user identity, SSO and budgeting controls that start on the Teams plan.
What should you use instead of OpenHands?
Claude Code is a stronger pick if you want Anthropic's own terminal agent with tighter Claude Sonnet integration and don't need to self-host. OpenAI Codex suits teams already invested in the ChatGPT ecosystem who want a managed, cloud-only coding agent instead of open-source infrastructure to run themselves. Cline is worth choosing if you want the same bring-your-own-model freedom packaged as a lightweight VS Code extension rather than a full deployable platform.
OpenHands or Claude Code: which should you pick?
Claude Code is Anthropic's first-party terminal agent, tuned for Claude models and installed with no server to run. OpenHands is open-source and model-agnostic: it can run the same Claude models, or swap in GPT, Gemini, Qwen, or a locally hosted open-weight model, and it can be self-hosted so code never leaves your infrastructure. Go with Claude Code for the smoothest single-model setup; go with OpenHands when model flexibility or owning your own infrastructure matters more.
What does it take to start using OpenHands?
For the free tier, follow the Agent Canvas setup guide in the OpenHands docs, add your own LLM API key under Settings, and point it at a Git repository through one of its native provider integrations. For the hosted route, OpenHands Cloud at app.all-hands.dev lets you log in with GitHub, GitLab or Bitbucket and connect a repo.
About OpenHands
OpenHands is an open-source AI coding agent built by OpenHands, a company founded in 2024 as All Hands AI by Robert Brennan, Graham Neubig, and Xingyao Wang and based in Boston, Massachusetts. Unlike a single-vendor assistant such as Claude Code (built by Anthropic) or OpenAI Codex, OpenHands is model-agnostic: point it at GPT, Claude, Gemini, Qwen, DeepSeek, or a locally hosted open-weight model, and its MIT-licensed core will run entirely on a developer's own machine. The project has passed 9 million open-source downloads and 500 contributors, and the company has raised $23.8M across a 2024 seed round and a November 2025 Series A led by Menlo Ventures.
Read the full background
OpenHands's CodeAct agent takes a GitHub issue or task description and works it autonomously from inside a sandboxed terminal, file system, and browser it controls directly, iterating on its own until the change is ready for a human to review rather than pausing for approval at each step. The V1 Software Agent SDK splits that behavior into modular, reusable agent, tool, and workspace packages and supports agent-to-agent delegation, so one agent can hand off a subtask to another inside the same run.
OpenHands fits engineering teams that want to inspect or self-host their coding agent rather than depend on one vendor's managed service: companies in regulated industries that can't let proprietary code leave their own infrastructure, platform engineers assigning it well-scoped GitHub issues, and technical founders who want to swap in a cheaper or open-weight model without changing their tooling. It sits alongside other coding agents such as Cline and Gemini CLI (built on Google's models), but its Agent Control Plane and Enterprise tier add centralized oversight, SAML/SSO, role-based access, and spend controls for organizations rolling agents out past a single developer. Independent reviewers describe it as better suited to teams experimenting with autonomous coding than to teams wanting a fully polished, zero-setup experience, since output quality still depends heavily on whichever LLM is connected.
OpenHands ships as three tiers: a free, MIT-licensed Open Source build that runs locally with your own LLM key through a web GUI, terminal UI, or CLI; a free-to-start Cloud plan with bring-your-own-key or at-cost model access; and a custom-priced Enterprise tier sold as SaaS or self-hosted inside a customer's own VPC. See the pricing FAQ below for exact plan limits and costs.
OpenHands ships new Enterprise versions at a fast clip, roughly one every one to two weeks: a March 2026 update added a beta Planning Agent that separates a high-level Plan Mode, which drafts a PLAN.md file, from an execution Code Mode, and the August 2026 Enterprise 0.41.0 release set GLM 5.2 as the default hosted model. The most recent release, Enterprise 0.71.1 on September 23, 2026, overhauled budget management for teams and dropped SQLite support in favor of PostgreSQL-only installations. The company also runs its own public benchmark, the OpenHands Index, which scores connected LLMs across five categories through its own SDK; as of June 2026 the best-performing backend, claude-fable-5, reached 95.8 out of 100 on the Issue Resolution category.
Sources and guides
Official sources
- OpenHands official site Vendor
- GitHub repository GitHub
Autonomy, guardrails, integrations and pricing on this page are read from OpenHands's own documentation. The verdict is Hokai's reading of them, not the vendor's claim.
Hokai guides that mention OpenHands
Lists that include OpenHands
More AI Agents on Hokai
PromptlessAI teammate that drafts doc updates as pull requests when you ship.
QualGentAutonomous QA agent that tests iOS and Android apps like a human.
OpenCodeOpen source coding agent that works with any model, in terminal, desktop or IDE.Codex SecurityOpenAI's security agent that scans repos and drafts fixes.
Antigravity CLIGoogle's terminal agent for multi-file coding, subagents and browser checks.
Muse CodeMeta's multi-agent CLI that plans, codes, and validates changes across repos.