OpenAI Codex CLI vs Gemini CLI vs OpenHands
Side by side, as alternatives to Cursor. Rows that differ are highlighted; the strongest value in a row is marked. Unknown means no source yet.
OpenAI Codex CLIopenai/codex
Gemini CLIgoogle-gemini/gemini-cli
OpenHandsOpenHands/OpenHands
Cursorpaid product
Cost and coverage
Price
$0 software; you run it
$0 software; you run it
$0 software; you run it
Free plan with limited AI use; Plus $8/month, Premium $30/month, Super $100/month, Ultra $200/month; third-party models billed separately by usage source ↗
Code in an AI-first editor
No: No
No: No
No: No
Confirmed: Yes
Multi-line tab completion and next-edit prediction
No: No
No: No
No: No
Confirmed: Yes
Let an agent plan and edit across many files
Partial: Partly
Partial: Partly
Partial: Partly
Confirmed: Yes
Run terminal commands and tests from the agent
Partial: Partly
Partial: Partly
Partial: Partly
Confirmed: Yes
Ask questions about your whole codebase
No: No
Partial: Partly
No: No
Confirmed: Yes
Connect tools and data through MCP
Partial: Partly
Partial: Partly
Partial: Partly
Confirmed: Yes
Run agents in the cloud in the background
No: No
No: No
Partial: Partly
Confirmed: Yes
Jobs covered
3 of 7 jobs
4 of 7 jobs
4 of 7 jobs
All 7
Popularity and activity (from GitHub)
Stars
130,321Top
107,492
91,499
—
Gain this week
Unknown
Unknown
Unknown
—
Signal
Building history
Building history
Building history
—
Commits / 4 wk
892Top
82
345
—
Latest release
rust-v0.163.0 · 2026-10-15
v0.65.0 · 2026-10-20
v1.30.0 · 2026-10-22Top
—
Commits / 12 wk
3,490
194
740
—
People committing / 12 mo
377+
82+
54+
—
Issues opened / closed, 30 days
7371 / 1498
123 / 206
464 / 367
—
Facts, with sources on each project page
Runs on my machine?Differs
Confirmed: Yes, a command-line agent that runs on your computer; cloud models run on OpenAI's servers unless you use --oss
Confirmed: Yes, the CLI runs in your terminal; the Gemini models themselves run on Google's servers
Confirmed: Yes; Agent Canvas and the agent server run on your machine by default, or on your own Docker host, VM or Kubernetes cluster
Hosted by the vendor
Code licenceDiffers
Confirmed: Apache-2.0
Confirmed: Apache-2.0
Confirmed: MIT
—
Model weights licence
Confirmed: No separate model weights
Confirmed: No separate model weights
Confirmed: No separate model weights
—
Commercial useDiffers
Confirmed: Allowed under Apache-2.0
Confirmed: Allowed under Apache-2.0
Confirmed: Yes, MIT allows commercial use
—
Paid services needed?Differs
Partial: Not required with local models (--oss with Ollama or LM Studio) or the limited ChatGPT Free plan; paid ChatGPT plans and OpenAI API keys are optional
No: A paid service is required since June 18, 2026; the free Google-account tier moved to Antigravity CLI. No local model support
Partial: Not required with local models (LM Studio, Ollama, vLLM, SGLang); cloud model APIs and OpenHands Cloud are optional
Subscription
HardwareDiffers
Confirmed: 4 GB RAM minimum, 8 GB recommended; models run in OpenAI's cloud, so no GPU needed
Confirmed: 4GB+ RAM for short sessions, 16GB+ for large codebases; needs Node.js 20+ and an internet connection
Partial: No requirements given for Agent Canvas itself; the recommended local model needs a 24GB GPU or an Apple Silicon Mac with 64GB memory
—
Operating systemsDiffers
Confirmed: macOS, Linux and Windows (native PowerShell with a Windows sandbox, or WSL2)
Confirmed: Windows 11 24H2+, macOS 15+ and Ubuntu 20.04+
Confirmed: macOS, Linux and Windows (PowerShell or WSL2); the CLI and per-conversation Docker sandboxes need WSL on Windows
—
Setup effortDiffers
Confirmed: Easy; one-line install script, npm or Homebrew, then sign in with ChatGPT
Confirmed: Easy; run with npx or install with npm, Homebrew or MacPorts, then sign in or set an API key
Confirmed: Medium; install Node.js 24 and uv then the npm package, or run the Docker image, then add a model
—
Stars over time
Switch to Growth % to compare projects of different sizes fairly.
The chart appears after a few nightly checks.
Notes for a video or post
A plain summary with today's date and a link to the sources. Check each project page before you publish.
Want share images and a ready video description? Open a project's creator kit.
Numbers: GitHub, fetched nightly. Facts: read from each project's own documentation, with the source on its page. How we check. openai/codex ↗google-gemini/gemini-cli ↗OpenHands/OpenHands ↗