Agentic harness comparison
Jules vs OpenAI Codex
Compared on where each one runs, which editors it supports, how autonomous it is, and how it bills — the four things that actually decide Jules vs OpenAI Codex. Rankings reflect fit and trade-offs, not a single universal winner; partner links, when present, do not replace that methodology.
Details last verified October 25, 2026
Jules
Cloud async agent (web, API and CLI client) · Agentic
OpenAI
OpenAI Codex
Desktop, CLI, IDE and cloud · Agentic
| Specification | Jules | OpenAI Codex |
|---|---|---|
| Vendor | ||
| Vendor | OpenAI | |
| Runs in | ||
| Runs in | Cloud async agent (web, API and CLI client) | Desktop, CLI, IDE and cloud |
| Editor support | ||
| Editor support | Browser workspace, GitHub integration, Jules API, Jules Tools CLI (cloud-session client) | Terminal beside any editor, VS Code / Cursor / Windsurf, JetBrains, Xcode, Codex desktop, Cloud |
| Autonomy | ||
| Autonomy | Agentic | Agentic |
| Pricing model | ||
| Pricing model | Hybrid | Hybrid |
| Requires changing editor | ||
| Requires changing editor | No | No |
Dates refer to the linked source’s publication or capture unless it explicitly states when the benchmark was measured. Values we could not confirm are shown as “Not verified” rather than estimated.
Jules
Google's Jules is a public-beta asynchronous coding agent that clones a GitHub repository into a short-lived Ubuntu cloud VM, develops a plan and code changes, and can produce a reviewable pull request. Web, API and CLI-client workflows have different approval behavior.
Best for
- Delegated async coding
- Open-source backlog triage
- Eligible individual Gmail users evaluating Google AI plans
Watch out
Free permits 15 tasks per rolling 24 hours/3 concurrent; Google AI Pro 100/15; Ultra 300/60. Paid upgrades currently require individual @gmail.com accounts and users must be 18+. Family users have separate task allowances. Free lists Gemini 2.5 Pro and paid tiers newer access starting with Gemini 3 Pro; exact regional subscription price is unverified. Web docs say plans eventually auto-approve on a timer if you leave; API sessions auto-approve plans by default unless requirePlanApproval is set. API PR creation is opt-in, and keys require the web GitHub-app setup. The VM has internet access. Google's no-training statement is scoped to private repository content, not universal zero retention. Limits block new tasks while existing tasks remain accessible. The earlier 51.8% secondary benchmark was removed because its original run configuration was not established.
OpenAI Codex
OpenAI's coding agent is available in desktop, terminal, supported IDE and cloud surfaces. ChatGPT sign-in uses plan credits and workspace policy; API-key sign-in uses separate Platform billing and organisation policy.
Best for
- Delegated coding tasks
- Existing OpenAI commitments
Watch out
ChatGPT sign-in uses Codex plan credits, whose rates differ from API prices. Do not estimate subscription task allowances from Platform token tariffs. Standard GPT-6.1 Sol rates are 50 input / 2.5 cached input / 250 output credits per million tokens; no separate cache-write charge. Plus is $20/month; Pro tiers are $100/$200/$500; Business Standard is $20/user/month with annual billing (two-user minimum), $25 monthly; Enterprise is negotiated. Pro has no five-hour limit, while Plus/Standard Business local-message estimates are not fixed caps; weekly limits and shared Work/Codex usage still apply. Fast mode bills at 2.5x included subscription credits or 2x purchased/pay-as-you-go credits; Ultrafast for GPT-6 Astra and GPT-6.1 Sol is 8x or 6x respectively and requires eligible access such as Pro 500. These are billing multipliers, not speed guarantees (https://learn.chatgpt.com/docs/pricing; https://learn.chatgpt.com/codex/agent-configuration/speed). GPT-6.1 Sol access is rolling out on Plus/Pro/Business/Enterprise/Edu; Enterprise/Edu admins must enable it. Free/Go desktop Luna access is rollout-dependent. Model availability varies by account, surface and client version (https://learn.chatgpt.com/docs/models). API-key sign-in supports local desktop/CLI/IDE work and uses Platform billing; cloud features such as GitHub review and Slack require ChatGPT sign-in (https://learn.chatgpt.com/docs/auth). GPT-5.4/mini retired from ChatGPT-authenticated Codex on August 31; GPT-5.5 has a documented retirement effective October 14, 2026 at noon PDT (19:00 UTC) for ChatGPT-authenticated Codex. API and API-key-authenticated Codex are unaffected; this review did not test an account. Pick a currently supported Sol/Luna replacement. Linux/WSL2 uses bubblewrap with a user-namespace-dependent bundled fallback; native Windows has its own sandbox and WSL1 is unsupported from CLI v0.115.0 (https://learn.chatgpt.com/docs/sandboxing).
Common questions
Jules vs OpenAI Codex
Answered from the verified figures on this page rather than general guidance.
Do I have to change editor to use Jules or OpenAI Codex?
How are Jules and OpenAI Codex priced?
Is Jules or OpenAI Codex better for large refactors?
Should I run Jules and OpenAI Codex together?
Will my project rules transfer between Jules and OpenAI Codex?
Is Codex still billed per seat or per message?
Do coding agents replace code review?
Should we harden MCP before rolling out a coding agent?
Next step
Which one fits
Autonomy and editor constraints decide this far more reliably than a benchmark score.
If most of your week is small, frequent edits, an assistive workflow beats delegation and an agentic harness sits idle while billing. If you regularly hand over work spanning many files, the reverse is true.
Where one tool requires changing editor and the other does not, that difference usually outweighs any capability gap across a team — rollouts fail on adoption, not capability.
“Open vendor” links use a first-party redirect for click counting. They are not commissionable partner placements unless a programme is active for that vendor. Affiliate disclosure.