GILD verdict
Choose Codex when OpenAI's models cover the work and engineers want the agent in their editor and desktop app, with usage included in ChatGPT plans.
Choose OpenHands when model choice or self-hosting decides the purchase. Teams that want both can run Codex inside Agent Canvas through ACP and compare the two agents on the same tasks and budget.
Buy on the model and the bill. On GPT-5.5 the two agents tie on All Hands AI's SWE-bench Verified run, and on GPT-5 Codex CLI led by about 6 points on the independent Terminal-Bench 2.0. Codex runs OpenAI models inside ChatGPT plans. OpenHands runs any model you connect, charges nothing for the open-source agent, and can be self-hosted end to end. Each ships an open-source local agent and a hosted service that turns tasks into pull requests.
Choosing between OpenHands and Codex
Choose OpenHands if
- You want models from several providers, open-weight models, or existing Bedrock, Azure, or Vertex AI agreements behind one agent.
- You want to self-host the whole system, including the agent server and sandboxes, in your own VPC.
- You want one control center that runs OpenHands alongside Codex, Claude Code, and Gemini CLI.
Choose Codex if
- You want model usage inside the seat: ChatGPT Business includes Codex at US$20 per user per month billed annually.
- Engineers want the agent in an IDE extension and the ChatGPT desktop app as well as the terminal.
- You want an OS-level sandbox on every laptop without running Docker.
Side by side
| Dimension | OpenHands | Codex |
|---|---|---|
| Open source | Agent Canvas, Software Agent SDK, and CLI under MIT; Enterprise is commercial | Codex CLI and SDK are open source on GitHub; the IDE extension and Codex Cloud are not |
| Models | Any model LiteLLM supports, including Claude, GPT, Gemini, and open-weight models served locally | OpenAI models through ChatGPT plans; the CLI can also point at custom providers, including Ollama and LM Studio, in config.toml |
| Individual priceList prices in US dollars. ChatGPT plans include usage. OpenHands bills model usage separately, so totals are not directly comparable. | Agent Canvas, the CLI, and the SDK: US$0 plus model usage. The pricing page lists no individual Cloud plan | Included in ChatGPT Plus: US$20 per month; Pro: US$100, US$200, or US$500 per month |
| Team price | Teams: US$500 a month for up to 10 users, self-hosted, with a 30-day trial and 5,000 sessions a month; Enterprise by quote | Included in ChatGPT Business: US$20 per user per month billed annually (2+ users), or US$25 billed monthly |
| Local sandbox | A Docker container by default; the process sandbox has no isolation | Seatbelt on macOS, bubblewrap on Linux and WSL2, and a native sandbox on Windows |
| Hosted and self-hosted | OpenHands Cloud, or Enterprise self-hosted on Kubernetes in your VPC | Codex Cloud tasks in cloud environments configured per repository, with internet access set per environment |
| SWE-bench Verified on GPT-5.5All Hands AI, which makes OpenHands, made and published both runs in the OpenHands Index, on its own harness and SWE-bench Verified's test split. | 78.2% at US$1.52 a task, OpenHands agent (SDK v1.18.1), submitted May 4, 202613 | 78.0% at US$1.17 a task, Codex run through ACP in the same harness, submitted May 4, 202614 |
| Terminal-Bench 2.0 on GPT-5Stanford University and Laude Institute run Terminal-Bench, independent of both vendors. Codex CLI also led on GPT-5-Mini, 31.9% to 29.2%. | 43.8% ±3.0, entered November 2, 202515 | 49.6% ±2.9 for Codex CLI, entered November 4, 202515 |
| Pull request work | Label an issue openhands or mention @openhands; a PR review workflow runs in GitHub Actions | @codex review on connected GitHub repositories, and the Codex GitHub Action |
What practitioners report, side by side
Each product's figures come from its own practitioners, for the group you pick.
Overall signal
All practitioners rate Codex +44, 29 points above OpenHands at +15. Most of the gap is in praise, with 59 of Codex's 106 sources positive against 21 of OpenHands's 52. Each has 12 negative.
The score counts the authors from all practitioners that take a side, one vote each with their latest view, and is how far positive votes lead negative ones, per 100 votes, so OpenHands's +15 rests on votes split 18 positive, 18 mixed and 11 negative and Codex's +44 on 55, 30 and 12.
A lead of a fifth of the votes reads positive and a lead of half strong positive, with the same lines below zero. The dots show each source, positive, mixed, negative. Use a task pilot to measure accepted work and review time.
Footprint on the Agentic Software Development stack
They differ on two of 12 layers. On Codebase context, OpenHands supports it and Codex is core. On Application security, OpenHands supports it and Codex touches it. Coverage is each product's placement on this stack, as its product page shows it. Core is a job the product owns; Supports and Touches are jobs it helps with. Coverage says nothing about how well a product does them.
| Layer | OpenHands | Codex |
|---|---|---|
| Build | ||
| Codebase context | Supports | Core |
| Validate | ||
| Application security | Supports | Touches |
The 10 layers where they match
| Layer | OpenHands | Codex |
|---|---|---|
| Plan | ||
| Work planning | Supports | Supports |
| Build | ||
| Coding | Core | Core |
| Agent runtime | Core | Core |
| Tool execution | Core | Core |
| Validate | ||
| Debugging | Supports | Supports |
| Testing | Supports | Supports |
| Code review | Supports | Supports |
| Ship | ||
| Source control | Touches | Touches |
| CI/CD | Touches | Touches |
| Operate | ||
| Observability | Touches | Touches |
What practitioners praise
Among all practitioners, Codex alone has six praise themes that reach 5 sources from 3 authors. Code Quality and Pull Request Review lead, at 27 of 33 and 12 of 12. Of the two themes both carry, Stability differs most. It is praised in 2 of the 5 OpenHands sources that discuss it, against 1 of 6 for Codex. OpenHands alone has four praise themes that reach 5 sources from 3 authors. End-to-end execution and Model choice lead, at 17 of 17 and 14 of 14.
| Praised for | OpenHands | Codex |
|---|---|---|
| Developer experience | 9 of 18 | 20 of 36 |
| Stability | 2 of 5 | 1 of 6 |
| End-to-end execution | 17 of 17 | Not a theme in its sources |
| Model choice | 14 of 14 | Not a theme in its sources |
| Open source & self-hosting | 9 of 10 | Not a theme in its sources |
| Async delegation | 4 of 6 | Not a theme in its sources |
| Code Quality | Not a theme in its sources | 27 of 33 |
| Pull Request Review | Not a theme in its sources | 12 of 12 |
| Long-Running Autonomy | Not a theme in its sources | 12 of 17 |
| Value for Money | Not a theme in its sources | 11 of 15 |
| Codebase Understanding | Not a theme in its sources | 10 of 11 |
| Fast Runs | Not a theme in its sources | 6 of 12 |
Each count is the sources that praise a theme, or raise it as a concern, out of those that discuss it. A row pairs the two products when both carry a theme under the same or an equivalent name.
A count ending in "so far" has under 5 sources from 3 authors among all practitioners. "Not a theme in its sources" means GILD found no theme of that name in that product's sources.
What practitioners raise as concerns
Among all practitioners, Codex alone has five concern themes that reach 5 sources from 3 authors. Inconsistent Results and Speed lead, at 15 of 16 and 13 of 22. Both carry Usage and cost. It is raised in 9 of the 14 OpenHands sources that discuss it, against 22 of 39 for Codex. OpenHands alone has four concern themes that reach 5 sources from 3 authors. Unpredictable task success and Setup & runtime friction lead, at 18 of 23 and 17 of 21.
| Raised as a concern | OpenHands | Codex |
|---|---|---|
| Usage and cost | 9 of 14 | 22 of 39 |
| Unpredictable task success | 18 of 23 | Not a theme in its sources |
| Setup & runtime friction | 17 of 21 | Not a theme in its sources |
| Security & access | 7 of 7 | Not a theme in its sources |
| Dependence on the model | 7 of 8 | Not a theme in its sources |
| Inconsistent Results | Not a theme in its sources | 15 of 16 |
| Speed | Not a theme in its sources | 13 of 22 |
| Requires Strong Review | Not a theme in its sources | 9 of 23 |
| Communication & Steering | Not a theme in its sources | 7 of 12 |
| Needs Close Steering | 3 of 4 so far | 6 of 9 |
22 sources
- OpenHands pricing
- OpenHands on GitHub: LICENSE
- OpenHands docs: LLM overview and model recommendations
- OpenHands docs: Sandbox overview
- OpenHands docs: Docker sandbox
- OpenHands docs: ACP agents in Agent Canvas
- OpenHands docs: GitHub integration
- OpenHands docs: PR review workflow
- OpenHands docs: OpenHands Enterprise
- OpenHands Index: GPT-5.5 results
- OpenHands Index: Codex through ACP on GPT-5.5
- OpenHands benchmarks: SWE-bench harness
- OpenHands Index: GPT-5.5 scores and cost per task (submitted May 8, 2026)
- OpenHands Index: Codex through ACP on GPT-5.5, scores and cost per task (submitted May 4, 2026)
- Snorkel AI: Terminal-Bench 2.0 leaderboard, Stanford University and Laude Institute (as of Oct 25, 2026)
- Codex docs: Pricing
- Codex docs: Sandboxing
- Codex docs: Open source
- Codex docs: Advanced configuration
- Codex docs: Cloud
- Codex docs: GitHub code review
- Codex docs: Codex GitHub Action