OpenHands
OpenHands is an open-source AI coding agent developed by All Hands AI, capable of autonomously writing code, fixing bugs, and releasing features through interaction with software environments.
OpenHands (formerly OpenDevin) is the leading open-source autonomous software-engineering agent, built by All Hands AI. Rather than autocompleting inside an editor, its agents write code, run shell commands, browse the web, and open pull requests end-to-end inside sandboxed Docker/Kubernetes environments. As of June 2026 it is the most-starred open-source coding agent (~76K GitHub stars), is fully model-agnostic, and is backed by an $18.8M Series A (Nov 2025, led by Madrona), making it the leading open alternative to proprietary agents like Devin and Claude Code.
- Built by
- All Hands AI
- Category
- autonomous SWE agent (open-source, model-agnostic)
- Open source
- Yes
- Models
- Model-agnostic via a LiteLLM provider layer (100+ providers). Supports Anthropic Claude, OpenAI GPT/Codex, Google Gemini, Mistral, DeepSeek, Qwen, plus local models via Ollama / llama.cpp and Bedrock/Vertex/Azure. Best results with frontier models (Claude Opus/Sonnet, GPT-5.2-codex). The free Cloud tier runs on a MiniMax model.
Available on
Pricing
Open-source core is free (MIT-licensed, self-host). OpenHands Cloud: sign up free with $20 in starter credits on a pay-as-you-go plan; the free Individual tier is capped at ~10 conversations/day with bring-your-own-key or at-cost (no-markup) usage of the OpenHands LLM provider. Optional Pro subscription is $20/month (covers cloud runtime compute, unlocks BYOK across providers and no-markup model usage). Enterprise (SaaS or self-hosted, multi-user RBAC) is custom-priced via sales.
Benchmarks
- SWE-bench Verified (CodeAct/SDK with Claude Sonnet 4.5, extended thinking)72.8%
- SWE-bench Verified (CodeAct/SDK with GPT-5, reasoning=high)68.8%
- SWE-bench Verified (CodeAct/SDK with Qwen3 Coder 480B A35B)65.2%
- SWE-bench Verified (CodeAct v2.1 + Claude 3.5 Sonnet, top open-source entry, Apr 2025)53.0%
Key capabilities
- ▸CodeAct architecture: the agent writes and executes Python code for actions (read files, run tests, apply patches) rather than JSON tool-calls, which outperforms schema tool-calling on multi-step tasks
- ▸Sandboxed execution in Docker or Kubernetes, with every action logged and reviewable
- ▸Four surfaces: Python SDK, CLI (pip install openhands-ai), local web GUI, and OpenHands Cloud
- ▸Native Git integrations (GitHub, GitLab, Bitbucket, Slack) - tag the agent on an issue/PR and it opens a branch
- ▸Planning Mode (beta, v1.6.0): switch between Plan and Code modes; agent generates a structured PLAN.md and asks clarifying questions
- ▸Kubernetes self-hosting with multi-user RBAC and audit logs for enterprise/regulated on-prem deployments
Strengths
- +Fully open-source (MIT core) and truly model-agnostic - no vendor lock-in, can run entirely locally for data privacy
- +Top-tier benchmark performance for an open agent when paired with a strong model (~72.8% SWE-bench Verified with Claude Sonnet 4.5 per the official SDK paper)
- +Multiple surfaces (SDK/CLI/GUI/Cloud) and fast release cadence; managed Cloud removes Docker setup
- +Enterprise-ready path: Kubernetes, RBAC, on-prem self-hosting for regulated industries
Limitations
- −Quality and cost depend heavily on the chosen LLM; running hard against frontier models gets expensive, and cheaper models degrade results sharply
- −Free Cloud Individual tier is constrained (~10 conversations/day); self-hosting can require non-trivial setup
- −No longer at the very top of the SWE-bench leaderboard - proprietary scaffolds (e.g. Augment Code) have pushed higher; V0->V1 architecture split (Nov 2025) means much older documentation describes a different codebase
Best for
Developers and engineering teams who want a powerful, self-hostable, model-agnostic autonomous coding agent without proprietary lock-in, especially those needing on-prem/data-residency control.
What's new
v1.6.0 (March 30, 2026) shipped Kubernetes self-hosting (multi-user RBAC) and a Planning Mode beta. In January 2026 the team launched the OpenHands Index, a broader evaluation spanning issue resolution (SWE-bench Verified), greenfield development (commit0), frontend (SWE-bench Multimodal verified), testing (SWT-Bench) and info-gathering (GAIA). Agent Canvas - a self-hostable control center that orchestrates OpenHands, Claude Code, Codex, Gemini or any ACP-compatible agent - is a newer parallel initiative moving into the flagship repo.
Signal Radar
Five-axis snapshot of this entity's footprint
Mentions × Lab Attention
Weekly mentions (solid) and average article relevance (dotted)
Timeline
No timeline events recorded yet.
Recent Articles
2Aider vs Claude Code: When to Use Each for Terminal-First Development in 2026
~Aider vs Claude Code: Aider wins on cost, undo, and local LLM support; Claude Code wins on agentic verification and enterprise security. Choose based
100 relevanceHarbor Adds LangSmith Sandbox Support, Making Agent Eval Backends Swappable
~Harbor, an open-source agent-evaluation framework, now integrates LangSmith sandboxes. This allows users to run the same eval across multiple provider
78 relevance
Predictions
No predictions linked to this entity.
AI Discoveries
No AI agent discoveries for this entity.
Sentiment History
| Week | Avg Sentiment | Mentions |
|---|---|---|
| 2026-W24 | -0.30 | 1 |
| 2026-W28 | 0.00 | 2 |