Skip to main content
OpenHands+15 Mixed

GILD Review

GILD's call and the buying facts behind it.

From GILD research

GILD analysis

OpenHands Review

Choose OpenHands when your team wants to decide which model the agent calls and where it runs, with no per-seat agent fee on the open-source path. It runs on any model LiteLLM supports, under the MIT license, and can run entirely inside your own infrastructure. The cost is ownership: your team configures sandboxes and secrets and pays model providers directly.

  • AI Coding Agents
  • Agentic Development

The decision

Best fit
Teams that must keep code, model calls, and sandboxes on their own infrastructure, or that want to use existing model contracts.
Consider another workflow
If you want a managed agent with a published per-seat plan and no sandbox to operate, Devin or Codex fits better.
Adoption condition
A Docker or managed sandbox, a strong model with a budget per conversation, repository setup scripts and AGENTS.md, and human review before merge.

Basis: OpenHands documentation, pricing, licenses, and published OpenHands Index results. On the independent Terminal-Bench 2.0 leaderboard, OpenHands with Claude Opus 4.5 scored 51.9%, level with Claude Code's 52.1% on the same model.

Buying & deployment factsWhat engineering leaders should verify before a broader rollout.
How you buy it

Free, MIT

Agent Canvas, the CLI, and the SDK are free. Teams is US$500 a month for up to 10 users.

Source 1, openhands.dev, Pricing: PricingSource 11, GitHub, Agent Canvas on GitHub: Agent Canvas on GitHubSource 12, GitHub, Software Agent SDK on GitHub: Software Agent SDK on GitHubSource 13, GitHub, OpenHands CLI on GitHub: OpenHands CLI on GitHub

Details

Agent Canvas, the CLI, and the SDK are free under MIT. Teams is self-hosted at US$500 a month for up to 10 users, plus US$50 a month per extra seat, with SAML SSO, RBAC, and a 30-day trial, capped at 5,000 sessions a month. Enterprise has custom pricing and unlimited seats. The pricing page lists no OpenHands Cloud plan.

Cost drivers

No markup

Model usage at vendor rates is the cost. Cloud caps spend per conversation and per month.

Source 2, All Hands AI, LLM overview: LLM overviewSource 5, All Hands AI, OpenHands LLM provider: OpenHands LLM providerSource 9, All Hands AI, Application settings: Application settingsSource 8, All Hands AI, Organization budgets: Organization budgets

Details

Model usage is the main cost, through your own provider key or the OpenHands provider at vendor rates with no markup. Claude Sonnet 4.5 costs US$3 per million input tokens and US$15 per million output tokens. Cloud caps spend with a maximum budget per conversation and a monthly organization budget.

Where it runs

Local or your VPC

A Docker sandbox, your own VM or Kubernetes, or OpenHands Cloud's managed sandboxes.

Source 3, All Hands AI, Sandbox overview: Sandbox overviewSource 6, docs.openhands.dev, Kubernetes (Helm) backend: Kubernetes (Helm) backendSource 7, docs.openhands.dev, Cloud backend: Cloud backendSource 4, All Hands AI, Enterprise vs. open source: Enterprise vs. open source

Details

Locally in a Docker sandbox or as an unisolated process, or on your own VM, Docker, or Kubernetes backend. Hosted in OpenHands Cloud's managed sandboxes, or as self-hosted OpenHands Enterprise on Kubernetes.

What to verify

Write access

The Cloud GitHub app asks to write Actions, contents, and workflows. Isolation varies by backend.

Source 11, GitHub, Agent Canvas on GitHub: Agent Canvas on GitHubSource 12, GitHub, Software Agent SDK on GitHub: Software Agent SDK on GitHubSource 13, GitHub, OpenHands CLI on GitHub: OpenHands CLI on GitHubSource 4, All Hands AI, Enterprise vs. open source: Enterprise vs. open sourceSource 3, All Hands AI, Sandbox overview: Sandbox overviewSource 10, All Hands AI, GitHub integration (Cloud): GitHub integration (Cloud)

Details

Agent Canvas, the SDK, and the CLI are MIT-licensed, and Enterprise is commercial. Isolation depends on the sandbox each backend uses. The Cloud GitHub app requests write access to Actions, contents, and workflows.

Sourced from OpenHands documentation