Best for · scenario map · 15 Sep 2026
Best cloud coding agents
Most 2026 lists compare editors, terminals, and a single vendor agent. The useful split is the job: isolated cloud, live or unattended steer, a pull request, and what you can show a reviewer when the run stops.
Facts checked 15 September 2026. Prices and limits move. Read the citations.
Rush Cloud is our product. It is one row: the multi-harness runtime. It is not 'the best agent.'
Pick the engine first (Claude Code, Codex, Copilot, Devin, Cursor, Gemini, Amp, Factory). Then pick the runtime.
Read this first
The useful distinction
Most 2026 lists mix editors, terminals, and a single vendor agent. The useful split is the job: isolated cloud, a pull request, and what you can show a reviewer when the run stops.
Vendor merge rates are not a shared benchmark. Cognition's 67% (14 Nov 2025) and Cursor's 60% internal (2 Sep 2026) measure each vendor using itself.
Routing
If you… use… not if
| If you | Use | Not if |
|---|---|---|
| Already run Claude Code, Codex, or OpenCode | Rush Cloud, or the vendor web for a single harness | You wanted a new agent personality |
| Want an autonomous vendor engineer on tickets | Devin or Factory Droids | You must bring your own model keys (Devin) or you need published rate limits (Factory) |
| Want the editor you already pay for, off the laptop | Cursor Cloud Agents | You are on Hobby |
| The work is a GitHub issue in a well-tested repo | Copilot cloud agent or Jules | You need more than 59 minutes (Copilot) or a non-GitHub remote (both) |
Sourced numbers
What we can chart
Self-serve starting price (individual, USD / month)
List prices from vendor pages retrieved 15 September 2026. Rush has no seat fee: the free plan includes 5 agent-hours a month and Pro is $20, so the bar is the seat, not the bill.
Jules daily task caps (rolling 24 hours)
From jules.google/docs/usage-limits, retrieved 15 September 2026.
Amp orb standard hourly prices
From ampcode.com/docs/orbs/sizes-and-costs, retrieved 15 September 2026. Billed by the minute. Paused orbs cost nothing.
Landscape
The field
Prices and limits from vendor pages retrieved 15 September 2026. Check the source before you buy.
| Tool | Job | What runs | Isolation | Proof | Price | Taps out |
|---|---|---|---|---|---|---|
| Rush Cloud | Run the harness you already use, in an isolated team session | Claude Code, Codex, or OpenCode | One sandbox per run | PR + trajectory, or needs_review | Free 5 agent-hours / month, then $20 Pro | Three harnesses. us-west-2. Customer VPC is roadmap. |
| Devin | Cognition's autonomous engineer on a ticket | Devin Brain + Devbox | Isolated VM per session | PR on your SCM | Pro $20 / mo; Teams $80 min | No BYO model keys. Opaque quotas. |
| Cursor Cloud Agents | Cursor Agent, unattended, in a cloud VM | Cursor Agent | Firecracker microVM | Draft PR + artifacts | Paid Cursor plan, from $20 / mo | Needs Cursor. Commands auto-run. |
| GitHub Copilot cloud agent | Issue-to-PR inside GitHub Actions | Copilot cloud agent | Ephemeral Actions VM | Draft PR + session logs | Paid Copilot, from $10 / mo | 59-minute cap. GitHub repos only. |
| Claude Code on the web | Claude Code, hosted by Anthropic | Claude Code | Anthropic VM (~4 vCPU / 16 GB) | GitHub branch + PR | Claude Pro $20 / mo and up | Research preview. Claude and GitHub only. |
| Codex cloud | OpenAI Codex in a ChatGPT cloud box | Codex | Container; agent phase offline by default | Diff + PR on GitHub/GitLab | ChatGPT Plus $20 / mo and up | API-key path has no cloud. |
| Factory Droids | Factory agent on a persistent computer | Droid | Managed 4 CPU / 8 GB computer | PR from Slack or the app | Pro $20; managed computers from Plus $100 | Unpublished rate limits. Own harness. |
| Google Jules | Async Gemini agent on a GitHub issue | Jules / Gemini | Fresh Ubuntu VM per task | GitHub PR | 15 tasks / day free; paid via Google AI | GitHub only. Individual Gmail plans. |
| Amp orbs | Amp agent on a remote machine | Amp | Fresh orb per thread | You push; gh pr create via AGENTS.md | Hobby free; Individual $20 + orb hours | Own harness. GHES unsupported. |
Depth
In depth
Free 5 agent-hours a month, then $20 Pro. 15 September 2026.
Rush Cloud
Multi-harness teams who want a Project and a trajectory.
The hole in the 2026 SERP is a runtime that is not also an editor and not also a model. That is the Rush row.
Taps out. Three harnesses. us-west-2.
15 September 2026
The vendor clouds
Staying inside one company.
Devin, Cursor Cloud Agents, Copilot cloud agent, Claude Code on the web, Codex cloud, Factory, Jules, and Amp each sell a complete loop. Read the vs pages before you switch the engine and the runtime on the same day.
Taps out. Each page in this cluster names the tap-out.
Honest limits
Skip Rush if
- You need a team that already runs Claude Code, Codex, or OpenCode. Rush does not invent a fourth harness.
- Sandboxes are us-west-2. Customer VPC is roadmap. A signed compliance report is not something we publish today.
- Compute is metered. Sixty minutes are included per user per calendar month. The invoice bills sandbox time after that allowance.
- You will review the PR. Rush does not merge for you.
FAQ
Questions people actually ask
Can I use Rush and the vendor cloud side by side?
Yes. Many teams keep Claude Code on the web or Cursor Cloud Agents for a single-vendor loop and put Rush on the tasks that need a shared Project, a second harness, or a trajectory they can hand to a reviewer.
Does Rush replace Claude Code or Codex?
No. Rush runs those binaries. You keep the seat or the key. Rush is the runtime.
Where does the code go when I leave?
Anything pushed to your GitHub remote stays there. That is the off-ramp on every product in this set, including Rush.
Why is there no ranking?
A crown needs a shared benchmark. Vendor merge rates (Devin 67% in Nov 2025, Cursor 60% internal in Sep 2026) are not that benchmark. Scenario routing is.
Cluster
Keep reading
Citations
Sources
Every factual claim on this page traces to one of these URLs. Retrieval date is 15 September 2026 unless the source prints an earlier published date.
- [1] Rush, “Rush Cloud product requirements”, retrieved 2026-09-15. https://tryrush.ai/cloud
- [2] Cognition, “Devin's 2025 Performance Review”, published 2025-11-14. https://cognition.com/blog/devin-annual-performance-review-2025
- [3] Cursor, “Self-hosted machines for Cloud Agents”, published 2026-09-02. https://cursor.com/blog/self-hosted-machines
- [4] GitHub Docs, “About GitHub Copilot cloud agent”, retrieved 2026-09-15. https://docs.github.com/en/copilot/concepts/agents/cloud-agent/about-cloud-agent
- [5] Anthropic Docs, “Use Claude Code on the web”, retrieved 2026-09-15. https://code.claude.com/docs/en/claude-code-on-the-web
- [6] OpenAI / ChatGPT Learn, “Codex cloud”, retrieved 2026-09-15. https://learn.chatgpt.com/docs/cloud
- [7] Start Debugging, “Claude Code vs Cursor vs Copilot Agent Mode: Where Each Wins in 2026”, published 2026-06-03. https://startdebugging.net/2026/06/claude-code-vs-cursor-vs-copilot-agent-mode-where-each-wins/
Next
Run the next Task on Rush Cloud
Connect a repository. Pick Claude Code, Codex, or OpenCode. Review the PR.