What is Claude Code Projects and how is it different from Codex, Copilot and Jules?

Claude Code Projects, launched September 17, 2026, adds a persistent coordinator conversation above multiple Claude Code cloud sessions: you brief it in natural language, it splits work into parallel worker threads, each on its own branch returning pull requests, and it keeps a shared project memory of decisions. OpenAI Codex offers a multi-agent workspace and cloud sandboxes across every ChatGPT plan, with the Agents API in public beta. GitHub Copilot's cloud agent is repository-native — tied to issues, branches, PRs and CI — and relaunches as a unified Copilot experience from September 28, 2026. Google Jules is an asynchronous GitHub-issue solver with repository-level memory. All four run agents in parallel; they differ on where the control plane sits.

How much do Claude Code Projects, Codex, Copilot coding agent and Jules cost?

Claude Code Projects has no separate price: it draws on a Claude Pro ($20/month) or Max ($100 or $200/month) subscription, and parallel threads consume usage faster. Codex is included in all ChatGPT plans, including Free, with higher limits on Plus ($20) and Pro ($200). GitHub Copilot's cloud agent is billed under Copilot plans with premium-request metering. Jules has a free tier of 15 tasks per day, an individual plan at $20/month (raised from $15 in March 2026) and a Team plan at $50 per user per month.

Which coding agent manager is best for a team in 2026?

If your repository is already the control plane — issues, PRs, CI — GitHub Copilot's cloud agent is the least friction. If you want one long-running conversation to coordinate work across several repos and steer it from a phone, Claude Code Projects is the only product with an explicit coordinator plus shared memory, though it is beta, cloud-only and Pro/Max-only today. Codex is the widest-available option and the one to pick if you need a free tier or an API (Agents API beta). Jules is the cheapest way to clear a backlog of small GitHub issues asynchronously.

Can Claude Code Projects access my local files?

Not yet. As of September 19, 2026, Claude Code Project worker threads run only in the cloud and cannot reach files or tools on your computer or services on your internal network. Anthropic says local support is coming 'very soon.' Until then, projects that depend on private packages, internal services or local tooling should use Claude Code's local sessions or a competitor with local execution.

Quick Answer

Claude Code Projects vs Codex vs Copilot Agent vs Jules

Published: September 19, 2026

The layer above the coding agent

By September 2026 every major vendor can run coding agents asynchronously and in parallel. The competition has moved up a level: who coordinates those agents, what remembers their decisions, and where a developer goes to manage a body of work that outlives any single session. Anthropic’s Claude Code Projects (September 17, 2026) is the newest answer. Here is how it stacks up against OpenAI’s Codex, GitHub’s Copilot cloud agent and Google’s Jules.

	Claude Code Projects	OpenAI Codex	GitHub Copilot cloud agent	Google Jules
Launched	Sep 17, 2026 (beta)	2025; Agents API beta Sep 2026	2025; unified relaunch ≥ Sep 28, 2026	2025; $20 tier since Mar 2026
Control plane	One coordinator conversation	Multi-agent workspace (Codex app)	The repository (issues, PRs, CI)	GitHub issue labelled `jules`
Parallel work	Coordinator spawns worker threads	Parallel sessions/sub-agents	Multiple isolated sessions	Multiple async tasks
Memory	Shared project memory across threads + library	Per-session; ChatGPT memory	Repo instructions; chat retained	Repository-level memory
Execution	Cloud only (local “very soon”)	Cloud sandboxes + local CLI/app	Cloud sandbox	Cloud VMs
Output	One branch + PR per thread	Diffs, PRs, commands	PRs tied to issues	PRs
Models	Claude Opus 5 / Fable 5.1 / Sonnet 5	GPT-6 Astra, GPT-5.6 Sol	Opus/Sonnet 5, GPT-5.x, Gemini 3.6/3.7 Flash	Gemini only
Mobile steering	Yes	Codex mobile; ChatGPT app	GitHub Mobile (unified Sep 28)	Web
Price	Included in Claude Pro/Max; usage-metered	All ChatGPT plans incl. Free	Copilot plans + premium requests	Free 15 tasks/day; $20/mo; Team $50/user
Availability	Select Pro/Max cloud-session users; Team/Enterprise later	GA	GA	GA

Verified September 19, 2026. Pricing from vendor pages; Jules figures include its March 2026 increase.

Claude Code Projects — the coordinator

Anthropic’s launch line is “threads do the work, Claude directs it.” A Project has a coordinator above a set of worker threads; you keep talking to the coordinator as requirements change, and it decides whether a request belongs in an existing thread or a new one. Each worker is a full Claude Code session on its own branch, returning changes as standard pull requests. An Overview panel surfaces items needing your attention, and you can steer individual threads from a phone.

What is genuinely new is shared operational memory across threads — a release date moving, a feature being dropped, a rule about who must be consulted before touching a service — plus a library of user files and Claude-created artefacts. Each Project can have its own cloud environment, connectors, plugins, instructions and model/effort settings for coordinator and workers.

Limits today: it is a beta for select Pro and Max subscribers using cloud sessions; threads are cloud-only and cannot see local files or your internal network; and because a Project runs several full sessions at once it burns through usage limits faster. There is no separate price. (Do not confuse it with the older Claude Projects workspace from June 2024, which groups chats and files but does not coordinate work.)

OpenAI Codex — the widest reach

Codex is available on every ChatGPT plan, including Free, as a cloud agent in isolated sandboxes, a local CLI and a desktop app that operates as a multi-agent workspace for running several agents in parallel with isolated changes. GPT-6 Astra has been in Codex since early September 2026; GPT-5.5 retires from Codex on October 14, 2026. The Agents API entered public beta in September 2026, exposing the same harness — session management, context compression, error recovery, sub-agent delegation — to developers.

Strength: availability, price and an API. Weakness: no explicit coordinator with cross-session memory; you are the scheduler. Mid-September brought user reports of capacity errors when running many parallel sessions, and Codex users must be on 0.146.0+ for the Plugin4Shell fix.

GitHub Copilot cloud agent — the repository-native one

GitHub’s agent lives where the work already is: assign it an issue, it opens a PR, CI runs, reviewers comment, it iterates. From September 28, 2026 Copilot Chat on github.com, GitHub Mobile and the cloud agent merge into a single unified Copilot experience under one policy, with the cloud agent moving to a faster sandbox. GitHub said on September 16 that the agent runtime was rewritten in more than 800,000 lines of Rust, largely by AI agents. Model line-up churns fast: Opus/Sonnet 4.x and Gemini 3.1 Pro were retired September 1, replaced by Opus/Sonnet 5 and Gemini 3.6/3.7 Flash; more deprecations land October 19.

Strength: zero new control plane if GitHub already is yours; enterprise admin dashboards and usage APIs (September 17). Weakness: Copilot is the one agent still unpatched against Plugin4Shell as of September 19, 2026, and its coordination is per-issue, not per-initiative.

Google Jules — the backlog clearer

Jules clones your repo into a cloud VM, works an issue you labelled jules, runs tests and opens a PR; since September 2025 it reads review comments and pushes follow-up commits. It keeps repository-level memory so later tasks inherit preferences. Pricing: 15 free tasks a day, $20/month individual, $50 per user/month Team. It uses Gemini models only.

Strength: cheapest, simplest way to burn down small issues without touching your terminal. Weakness: no coordinator, no cross-repo work, and periodic doubts about its roadmap after it went unmentioned at Google I/O 2026 while Google pushes Antigravity as its developer platform.

Where each one puts the control plane

Anthropic: one continuing conversation coordinates many sessions, with shared memory — best for initiatives that span repos and weeks.
OpenAI: a workspace of parallel agents plus an API — best for reach and building your own orchestration.
GitHub: the repository — best when issues/PRs/CI are already how your team thinks.
Google: the issue — best for asynchronous, low-touch chores.

Recommendation

Solo developer on Claude Max with a long-running side project: Claude Code Projects, once you have access; it is the only tool here that remembers the plan for you.
Team living in GitHub: Copilot cloud agent after the September 28 relaunch — but hold marketplace plugins until Microsoft ships the Plugin4Shell fix.
Need a free tier, an API, or GPT-6 Astra: Codex.
Backlog of small issues, tiny budget: Jules.

Whichever you choose, the primitives are converging — parallel execution, isolated environments, memory, human intervention. The durable difference is where you go to think about the work.