What is the difference between Jules and Codex?
Jules and Codex cloud both take a coding task, run it on a remote machine against your GitHub repository and hand back changes, but Jules is a single-purpose async agent from Google, while Codex cloud is one surface of OpenAI's wider Codex product. Jules writes a plan for you to approve and then pushes a branch. Codex cloud runs a task from a reusable environment you set up, and you review the diff, then commit or open a pull request.
The bigger differences are around the edges: how environments are set up, what triggers a task, what a plan costs and which models you get. Everything below comes from the Jules docs and OpenAI's Codex cloud docs, read on 27 October 2026.
How does Google Jules work?
Google calls Jules "an experimental coding agent that helps you fix bugs, add documentation, and build new features." You sign in at jules.google.com, connect GitHub, and give it access to all or selected repositories.
A task follows the same steps every time:
- Pick a repository and branch, and write a scoped prompt.
- Click "Give me a plan." Jules "runs in a virtual machine where it clones your code, installs dependencies, and modifies files."
- Review the plan. The docs say "You can review and approve it before any code changes are made."
- Watch the activity feed or walk away. You can message Jules mid-task or pause it.
- Click Create branch to push the changes, then open the pull request on GitHub.
You can also start a task by adding the label jules to a GitHub issue. Per the running tasks page, Jules comments on the issue and links to a pull request when it is done.
The environment page says each VM runs Ubuntu Linux with Node.js, Bun, Python, Go, Java and Rust preinstalled. For anything unusual, you add commands under Initial Setup in the repository's Configuration tab, and Run and Snapshot saves the VM state so later tasks start faster. The FAQ says tasks run in a VM "with internet access" and that "Jules does not train on private repository content."
The changelog shows what was added in 2026: MCP connections to services such as Linear, Neon and Supabase, automatic fixes for failing GitHub Actions checks on Jules pull requests, and scheduled tasks you can now edit, pause and resume.
How do Codex cloud tasks work?
Codex cloud runs Codex coding tasks on OpenAI's machines. OpenAI's docs say "Each task has its own workspace and can keep working while your computer is asleep."
The core unit is a cloud environment: the GitHub repositories, tools and access settings each task starts from. Per the cloud environments page, Codex inspects your repositories and records an install script and a start skill, so you don't have to write setup by hand. Once published, every new task starts from that environment with its own workspace.
Internet access is set per environment. Turn on Enable all public internet access for full access, or leave it off and allow package managers or specific hosts under Additional allowed domains. Tokens go in environment variables or network secrets, where "Programs receive a placeholder; the proxy substitutes the real value for allowed destinations."
You can start a task from the ChatGPT web app, the desktop app, mobile or the terminal. The codex cloud command is marked experimental in the CLI's own help text:
# submit a task to an environment, with two best-of-N attempts
codex cloud exec --env ENV_ID --attempts 2 "fix the flaky date parsing test"
codex cloud list # recent cloud tasks
codex apply TASK_ID # apply a task's diff to your local checkout
On GitHub, @codex review on a pull request posts a review that, per OpenAI's GitHub integration page, flags only P0 and P1 issues. Other @codex mentions, like @codex fix the CI failures, start a legacy cloud chat with the pull request as context. The Codex cloud guide covers environments and reviews in more depth.
Jules vs Codex side by side
| Google Jules | Codex cloud | |
|---|---|---|
| Maker | OpenAI | |
| What it is | A standalone async coding agent | The cloud surface of Codex, next to the CLI, IDE extension and desktop app |
| Where you start a task | jules.google.com, the jules CLI, a jules label on a GitHub issue |
ChatGPT web, desktop and mobile, codex cloud exec, @codex mentions on GitHub |
| Repository host | GitHub | GitHub for cloud tasks (GitLab and GitHub Enterprise Server not yet supported); GitLab code review in beta |
| Environment setup | Optional setup script, saved as a snapshot | Published cloud environment with an install script Codex records for you |
| Internet in the task | VM has internet access | Set per environment: full public access, or package managers and allowed domains |
| Plan before code | Plan to approve before coding | Not described as a step in the cloud docs |
| Result | A branch you push with Create branch | A diff you commit, turn into a pull request, or pull locally with codex apply |
| Pull request review | Not described in the Jules docs | @codex review and automatic reviews |
| Project instructions | AGENTS.md at the repository root |
AGENTS.md files from the root down |
| Models | Gemini models | OpenAI models such as GPT-6.1 Sol and GPT-6 Luna |
| Entry price | No-cost plan | ChatGPT Plus at $20 a month |
How much do Jules and Codex cost?
Jules. The limits and plans page lists three tiers as of October 2026:
| Plan | Tasks per rolling 24 hours | Concurrent tasks | How you get it |
|---|---|---|---|
| Jules | 15 | 3 | No subscription needed |
| Jules in Pro | 100 | 15 | Google AI Pro |
| Jules in Ultra | 300 | 60 | Google AI Ultra |
The page prints no prices and says paid Jules plans come through a Google AI Plans subscription, "currently available only for individual Google Accounts (ending in @gmail.com)." On models, the official pages don't fully agree: the limits page lists Gemini 2.5 Pro for the base plan, while the changelog says Gemini 3 Flash became the base model for all tiers on 30 January 2026, and Gemini 3.1 Pro became the default for Pro users on 9 March 2026.
Codex. OpenAI's Codex pricing page lists, as of October 2026:
- Free ($0) and Go ($8 a month): GPT-6 Luna in the desktop app, subject to rollout, with no cloud features listed.
- Plus ($20 a month): Codex on the web, CLI, IDE extension and iOS, plus automatic code review and Slack.
- Pro (from $100 a month, with $100, $200 and $500 tiers): everything in Plus with more usage.
- Business ($20 per user a month billed annually, $25 monthly) and Enterprise and Edu (contact sales).
Cloud tasks share your plan's allowance with local work, and the page warns they "may use more of your allowance than local messages." Signing in with an API key gives the CLI and IDE extension but no cloud features.
Which should you pick?
Pick on what you already pay for and how you like to review work.
Jules tends to fit when:
- You want to try a cloud coding agent without a subscription.
- You like approving a written plan before any code changes.
- You want tasks to come from GitHub issues with a single label.
- You already have Google AI Pro or Ultra on a personal Gmail account.
Codex cloud tends to fit when:
- You already pay for ChatGPT Plus, Pro or Business.
- You want the same agent locally and in the cloud, and to pull a cloud diff into your checkout with
codex apply. - You want automatic pull request reviews on GitHub.
- You need tight control over network access and secrets per environment.
Both read AGENTS.md, so the cheapest way to compare is to give each the same small task on the same repository. See the AGENTS.md guide for what to put in that file, and the Jules guide or Codex CLI guide to set each one up. For GitHub's own cloud agent against Codex, read Copilot vs Codex.
Keep your task prompts in one place
Cloud agents work best with specific, reusable task prompts: "add tests for this module," "bump this dependency and fix the breakage." Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own, with fill-in blanks for the parts that change, and installs skills into Codex with one npx skills add command. To turn a prompt you keep reusing into a skill, the free SKILL.md generator runs in your browser.