OpenAI Codex Cloud vs Anthropic Claude Cowork (June 2026)
OpenAI Codex Cloud vs Anthropic Claude Cowork in June 2026
OpenAI’s June 11, 2026 Ona acquisition signals a Codex Cloud product coming. Anthropic’s Claude Cowork is already shipping with Claude Fable 5 baked in. Both bet that managed cloud agents — not local CLIs — are the real interface between developers and AI models. Here’s how the two strategies actually compare.
Last verified: June 13, 2026
TL;DR
| Dimension | Codex Cloud (post-Ona) | Claude Cowork |
|---|---|---|
| Status | Announced via Ona acquisition; not yet launched | Shipping, GA |
| Default model | GPT-5.5 (likely GPT-5.6 at launch) | Claude Fable 5 |
| Underlying compute | Ona’s microVM + Kubernetes infrastructure | Anthropic-managed |
| SDK | Codex SDK (extended) | Anthropic Agent SDK + Claude Code SDK |
| IDE surfaces | VS Code, Codex desktop | VS Code, Cursor, JetBrains, Claude desktop, web, iOS handoff |
| Chat surfaces | ChatGPT (planned) | Claude.ai, Slack |
| CI/CD | GitHub Actions, Azure DevOps (planned) | GitHub Actions, GitLab CI/CD |
| Sub-agents | Coming with Codex Cloud | Yes |
| Hooks & skills | Codex skills exist; cloud integration coming | Yes |
| Run inside customer cloud | Ona supported it; OpenAI continuation TBD | Limited |
| Best for | OpenAI-native enterprises waiting for the launch | Anyone shipping today |
The strategic context
Both companies decided at roughly the same time that the future of AI coding isn’t a local CLI — it’s an agent fleet running in cloud environments, coordinated by sub-agents, with hooks into your CI, your chat, your issue tracker. The CLI becomes the front-end; the cloud is the runtime.
Anthropic moved first. Claude Cowork (and its managed workflows) shipped in earlier 2026 and matured with the Fable 5 launch on June 9, 2026. OpenAI’s path was to buy Ona on June 11, 2026 rather than build from scratch. Both companies converge on the same product shape.
Anthropic Claude Cowork — what’s actually shipping
Default model: Claude Fable 5 after June 9, 2026.
Capabilities (June 2026):
- Server-side execution of Claude Code agents.
- Sub-agents for parallel multi-file work.
- Hooks (pre/post-execution callbacks).
- Skills (reusable agent procedures).
- Anthropic Agent SDK + Claude Code SDK.
- Integration with Cursor, Windsurf, VS Code, JetBrains.
- GitHub Actions and GitLab CI/CD support.
- Slack-native agent collaboration.
- Web and iOS handoff for agents you start on one device and continue on another.
Pricing: Bundled into Pro, Max, Team, and seat-Enterprise plans with metered Fable 5 usage. Paywall on subscription plans starts June 22, 2026 — see Claude Fable 5 paywall June 22.
Best for: Teams already on Claude Code who want to scale to fleet-of-agents and persistent cloud execution without changing model vendors.
OpenAI Codex Cloud — what’s coming via Ona
Status: Announced through Ona acquisition June 11, 2026. Not yet a launched product. Ona itself continues serving customers during the transition.
What Ona brings:
- Firecracker-style microVMs and Kubernetes orchestration.
- Persistent cloud development environments (Gitpod heritage).
- Ability to run inside customer’s own AWS / GCP / Azure account.
- Enterprise hardening: isolation, audit logs, network policies.
What Codex Cloud will likely add:
- Native Codex CLI ↔ cloud handoff (agent keeps running when laptop closes).
- Codex SDK integration for sub-agents in cloud-managed environments.
- ChatGPT-side agent surface for non-developers triggering Codex tasks.
- First-party billing via ChatGPT Business / Enterprise.
Pricing: Not announced. Realistic expectation: bundled credit allowance in ChatGPT Business + metered overage, mirroring Anthropic’s structure.
Best for: OpenAI-native enterprises building on Codex who can wait 3–6 months for first-party cloud agent execution.
Side-by-side: features developers actually care about
| Feature | Cowork | Codex Cloud (post-Ona) |
|---|---|---|
| Persistent agent state across sessions | Yes | Yes (Ona heritage) |
| Long-running agents (hours/days) | Yes | Yes |
| Sub-agents in parallel | Yes | Coming |
| Run inside customer VPC | Limited | Yes (Ona supported) |
| Multi-model support | Claude only | OpenAI only |
| SOC 2 / FedRAMP | Yes / partial | Ona had SOC 2; FedRAMP TBD |
| GPU access for inference inside agent | Limited | Likely via OpenAI infra |
| Open-source self-host | No | Gitpod heritage was OSS; uncertain post-OpenAI |
| Pricing transparency | Token-metered | TBD |
When each one wins
Pick Claude Cowork today when:
- You’re shipping a product that needs managed cloud agents now.
- Your team is already on Claude Code / Fable 5.
- You want fleet-of-agents with sub-agents and hooks shipping today.
- You value Slack-native + iOS handoff.
Wait for Codex Cloud when:
- Your product is locked into OpenAI (Codex SDK, GPT-5.5, ChatGPT integrations).
- You need to run agents inside your own AWS/GCP/Azure account.
- You can wait 3–6 months for first-party Codex Cloud GA.
- You want to bundle into existing ChatGPT Business / Enterprise billing.
Stay on a neutral sandbox (E2B, Modal) when:
- You build cross-model agents (Claude + GPT + Gemini in the same product).
- You don’t want vendor lock-in to either OpenAI or Anthropic.
- You need GPUs in the agent loop (Modal) or sub-150ms cold starts (E2B).
- See OpenAI Ona vs E2B vs Modal: AI agent sandbox platforms.
What this means for the AI coding market
- Managed agent products are the new battleground. Local CLIs are commoditized; the differentiator moves to cloud orchestration, sub-agents, skills, and CI/CD integration.
- Anthropic leads on shipping product; OpenAI leads on ecosystem reach. Cowork is more mature; Codex Cloud (when it lands) will reach 5M+ Codex weekly users instantly.
- Lock-in is the design goal. Both products are deliberately not multi-model — they exist to make switching cost-prohibitive.
- Cursor, Windsurf, GitHub Copilot all need a managed cloud story too. Expect Cursor Cloud, Windsurf Cloud, and a deeper GitHub Copilot CLI cloud offering by Q4 2026.
What to watch in Q3 2026
- Codex Cloud public preview announcement — likely tied to GPT-5.6 launch. See GPT-5.6 leaked features.
- Cowork pricing tier changes post-June 22 paywall transition.
- Microsoft’s response — GitHub Copilot CLI cloud agent runtime to compete with both. See Microsoft drops Claude Code.
- Cross-platform agent SDKs (LangGraph, CrewAI, Mastra) adapting to the new managed cloud reality.
Related coverage
- OpenAI acquires Ona (ex-Gitpod): Codex gets a cloud
- OpenAI Ona vs E2B vs Modal: AI agent sandboxes
- Claude Fable 5 paywall June 22, 2026: prepare your plan
- GPT-5.6 vs GPT-5.5 vs Claude Fable 5: which to use
Bottom line
Claude Cowork is shipping, Fable 5–powered, and the best managed cloud agent product available today. Codex Cloud will be the OpenAI-native equivalent within 3–6 months, with Ona-grade cloud infrastructure underneath. The question for buyers is the same as every other AI-coding question in mid-2026: are you betting on Anthropic, OpenAI, or staying neutral? Whichever way you bet, the local CLI era is ending faster than people realize.
Sources: OpenAI Ona acquisition announcement (June 11, 2026), Ona blog (Johannes Landgraf, June 11, 2026), Anthropic Claude Fable 5 launch (June 9, 2026), Augmentcode on Claude Code 131K stars (June 2026), Simon Willison, Bloomberg, CNBC, TechCrunch / TNW (June 11–12, 2026).