OpenHands is the open-source AI software engineer previously known as OpenDevin, maintained by All Hands AI. It reads, writes, runs, and debugs code inside a Docker sandbox, accessible via CLI, GUI, Python SDK, or OpenHands Cloud (docs.openhands.dev).
The repo carries 75K+ GitHub stars as of June 12, 2026. Model access is bring-your-own or provider-mediated: the stable claim is that OpenHands can be powered with Claude, GPT, or other LLMs through its configuration layer, not that one exact frontier model list will stay current.
Recent developments
- June 1, 2026: OpenHands pricing still lists three buying paths: free local Open Source, free SaaS or self-hosted VPC deployments. The old $20/mo Pro tier remains retired.
System Verdict
Pick OpenHands if autonomous software engineering at Devin-class capability is the goal and infrastructure ownership is acceptable. The MIT license, Docker sandbox, and GitHub PR integration make it production-credible for real tasks. Model flexibility is the moat against closed services.
Skip it if IDE autocomplete is the primary need. OpenHands delegates full tasks; it does not pair-program inline. Cursor or Claude Code fit in-editor workflows better.
Who pays which tier: is free with a 10 daily conversation cap and BYOK (or the OpenHands provider at cost, no markup). Enterprise covers SAML/SSO with custom pricing.
Key Facts
| Former name | OpenDevin (renamed 2024) |
| Maintainer | All Hands AI |
| License | MIT (core) · Enterprise directory is separate-licensed |
| GitHub stars | 75,500+ (June 1, 2026) |
| Interfaces | CLI · Local GUI · Python SDK · OpenHands Cloud |
| Sandbox | Docker container per task |
| Cloud Individual (free) | 10 daily conversations · BYOK or OpenHands provider at cost · API |
| Cloud Enterprise | SaaS or self-hosted in your VPC · SAML/SSO · unlimited concurrent runs · Large Codebase SDK · dedicated Slack channel · custom pricing |
| Model support | Claude, GPT, and other configurable LLMs; exact provider/model list depends on keys and deployment |
Every data point above was verified against vendor sources on 2026-06-12. See Sources.
What it actually is
An agent loop that wraps any LLM inside a controlled code-execution environment. Given a task, OpenHands plans the approach, edits files, runs tests, reads failures, iterates, and submits a pull request.
Deployment is your choice. CLI for terminal natives. Local GUI for browser-based work. Python SDK for custom agent automation. OpenHands Cloud for zero-setup evaluation.
The moat is low and the team knows it. The architecture is public, the model layer is swappable, and competing forks exist. Positioning rests on community velocity, benchmark leadership, and the cloud platform rather than protocol lock-in.
When to pick OpenHands
- Devin-class capability without the Devin bill. Self-hosting plus BYOK beats commercial Devin on price for most solo developers.
- Private infrastructure. Code must not leave the company network. Enterprise self-hosting inside your VPC gives full audit control.
- Model-agnostic coding agent. Benchmark your approved Claude, GPT, open-weight, or LiteLLM-compatible endpoint on the same task without rewriting the agent.
- Autonomous PR workflows. Feed an issue, get a reviewable pull request. Works on GitHub and GitLab natively.
- Research on agent architectures. Open evaluation infrastructure and Theory-of-Mind tooling for measuring agent performance.
When to pick something else
- IDE autocomplete and pair-programming: Cursor, Cline, or GitHub Copilot.
- Managed autonomous agent with zero setup: Devin (Cognition) for a commercial SLA.
- Terminal coding on Anthropic models: Claude Code CLI for deepest Claude integration.
- Visual agent workflow builder: Langflow or Relevance AI.
- Multi-agent team orchestration: CrewAI.
Pricing
| Tier | Cost | Notes |
|---|---|---|
| Self-hosted (MIT) | Free | Full features, BYO API key and infrastructure |
| Cloud Individual | Free | GitHub login, 10 daily conversations, BYOK or OpenHands provider at cost, no markup |
| Cloud Enterprise | Custom | SaaS or self-hosted VPC, SAML/SSO, unlimited concurrent runs, Large Codebase SDK, dedicated Slack channel |
| Model API costs | Usage-based | Depends on the configured provider/model, context size, and number of agent iterations |
Prices verified 2026-06-12 via OpenHands pricing. The $20/mo Pro tier was retired in May 2026; the current free Individual SaaS plan replaces it with a daily conversation cap and the same BYOK/at-cost model. Self-hosting under MIT remains fully free.
Against the alternatives
| OpenHands | Claude Code | Devin | |
|---|---|---|---|
| License | MIT (open-source) | Proprietary | Proprietary |
| Hosting | Self or Cloud | Anthropic cloud | Cognition cloud |
| Model choice | Any (BYOK) | Claude only | Cognition’s stack |
| Entry cost | Free (self-host) | $20/mo Pro | Paid per-task |
| Sandbox | Docker | Anthropic-managed | Cognition-managed |
| GitHub PR integration | Native | Native | Native |
| IDE autocomplete | No | No (CLI-first) | No |
| Best viewed as | Model-agnostic open-source agent | Anthropic-native CLI agent | Managed commercial SLA |
Failure modes
- No inline autocomplete. Task-delegation only. Pair-programming requires Cursor or Cline.
- Docker dependency for self-hosting. Full sandbox safety needs Docker. Lightweight mode drops isolation.
- Free Cloud caps at 10 conversations per day. Heavy users should self-host or move to Enterprise.
- Variable reliability on complex refactors. Large multi-file changes can exceed context limits or produce code requiring human fix-up.
- Low moat. Architecture is public; commercial services compete on reliability, not novelty.
- Slower than IDE tools for one-file edits. The plan-act-observe loop adds latency. Cline in VS Code wins on micro-edits.
- Enterprise directory has separate license terms. The enterprise extensions are not under MIT. Read the licensing before deploying.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility × Value × Moat × Longevity, unweighted average). Last verified 2026-06-12 against OpenHands documentation, OpenHands pricing, and the OpenHands GitHub repo.
FAQ
OpenHands vs OpenDevin: same project? Yes. OpenDevin was renamed OpenHands in 2024. All development and releases now ship under the OpenHands name, maintained by All Hands AI.
Is OpenHands free? Self-hosting is fully free under MIT. OpenHands Cloud offers a free Individual SaaS tier via GitHub login: 10 daily conversations, BYOK or the OpenHands provider at cost with no markup. The $20/mo Pro plan was retired in May 2026. Enterprise is custom-priced for SaaS or self-hosted VPC deployments with SAML/SSO, unlimited concurrent runs, and the Large Codebase SDK.
Does OpenHands work with Claude? Yes. OpenHands is model-agnostic. The project says the CLI can be powered with Claude, GPT, or other LLMs; exact model availability depends on the API keys, hosted provider, and deployment configuration.
OpenHands vs Devin? Both are autonomous software engineers that turn task descriptions into reviewable PRs. OpenHands is open-source, self-hostable, and model-agnostic. Devin is commercial, managed, and faster to onboard. OpenHands wins on price and control; Devin wins on reliability out of the box.
Can OpenHands open GitHub pull requests? Yes. Native GitHub and GitLab integration. It clones a repo, reads issues, implements a fix inside a Docker sandbox, and opens a PR for human review.
Sources
- OpenHands documentation: Architecture, deployment modes, agent loop
- OpenHands pricing: Cloud Individual and Enterprise tiers
- OpenHands GitHub: Source, stars, release history
- OpenHands Index announcement: January 2026 platform update
Related
- Category: AI Coding
