Skip to content
Promtime

Guide

Coding agents

How Claude Code, Codex, Cursor and the others differ

Updated 21 October 2026179 stories

In short

Claude Code, Codex and Cursor now differ less in what the agent can do and more in which models it ships with and how usage is billed. Codex runs OpenAI’s GPT-6 Astra and GPT-6.1 Sol, and GPT-6.1 Sol holds 500,000 tokens of context by default in Codex and ChatGPT Work (source). On the $200 Pro plan, Codex usage falls from 20x to 10x the Plus allowance on October 30. The new $500 Pro plan gives 25x Plus and Ultrafast for GPT-6 Astra (source). Ultrafast on GPT-6.1 Sol costs $12 per million input tokens and $60 per million output (source). In Claude Code, claude --cloud moves a session to an Anthropic-managed VM with no separate compute charge, but cloud sessions share rate limits with all your other Claude usage (source). Projects is still in beta and has let in everyone on its Pro and Max waitlist (source). Cursor now belongs to SpaceX. Its plans launched October 12 include Grok usage but no Claude, and on the $30 Premium plan every Claude request is billed at API list prices (source).

The latest changes: the Claude Code desktop app runs each parallel task in its own nested session (source). OpenAI changed how the Codex and ChatGPT desktop apps store state, which should cut the data kept on disk by 60% (source). That fix did not cover Crashpad dumps, and one user found 323 GB of them on a single Mac (source). Current Cursor subscribers keep their included Claude usage until they switch plans. Codex users on Pro 200 keep their current limits through October 29.

At a glance

AgentCompanyWhat matters nowPriceAs ofSource
Codex (ChatGPT Pro $200)OpenAIChatGPT Work and Codex usage is 20x the Plus allowance through October 29 and 10x from October 30; GPT-6 Pro in Chat drops from 200 to 100 messages per week; existing subscribers get 62,500 usage credits expiring December 31, 2026; the 5-hour limit won't come back$20030 September 2026Read
Codex (ChatGPT Pro $500)OpenAI25x the Plus allowance, plus Ultrafast for GPT-6 Astra with up to 8x faster token generation in Codex$50030 September 2026Read
Claude Code cloud sessionsAnthropicclaude --cloud runs the task on an Anthropic-managed VM that clones your branch from the GitHub remote; you steer it from claude.ai/code, the Claude app's Code tab or the Desktop app; claude --teleport brings it back to the terminalNo separate compute charge for the VM; cloud sessions share rate limits with all other Claude usage24 September 2026Read
Cursor PremiumCursor (owned by SpaceX)Grok comes from an included weekly pool; every Claude request is billed on top; the cheapest plan has no Claude; current subscribers keep their included Claude usage until they switch$30; Claude at API list prices13 October 2026Read
The X Subscription (Grok, Cursor, X)xAIOne pool of Grok model usage across the Grok apps, the Grok tab on X and Cursor; Premium bundles SuperGrok, Cursor Pro and X Premium+; third-party models in Cursor on Premium and above via top-ups at API list pricesPlus $8/month to Ultra $200; Premium $30/month13 October 2026Read
GPT-6.1 Sol in CodexOpenAIDefault context of 500,000 tokens in Codex and ChatGPT Work, up from 272k12 October 2026Read
GPT-6.1 Sol UltrafastOpenAI8x faster than Sol Standard; rolling out in ChatGPT Work, Codex and the API; OpenAI hasn't said whether Codex access requires Pro 500$12 per million input tokens, $60 per million output; Standard Sol $2/$108 October 2026Read
Claude Code ProjectsAnthropicBeta: one ongoing conversation in which Claude runs each task as its own thread in parallel; the whole Pro and Max waitlist has been let in, and new sign-ups get in as capacity allowsPro or Max plan9 October 2026Read
Claude Code desktopAnthropicEach parallel task runs in a nested session; Environments save a setup (folders, an SSH host or the cloud) with an optional worktree per folder; /btw opens side questions in their own tab14 October 2026Read
Codex cloud environmentsOpenAITasks keep running while your computer sleeps; each task starts from a published, reusable setup in its own workspace; you review from the web, mobile or desktop app29 September 2026Read
OpenAI DotsOpenAIThe primary Dot is included in the plan, runs 24/7 and uses no allowance on its own work; any Codex task it opens draws usage as usual; published terms promise inclusion only for the first month after launchIncluded in plan1 October 2026Read
Codex open models via BasetenOpenAI / BasetenGLM-5.3 Flash and Kimi K3 inside Codex for OpenAI enterprise customers; spend can count toward an existing OpenAI commitment; access is through a waitlist30 September 2026Read
Show all (18)
AgentCompanyWhat matters nowPriceAs ofSource
Codex Security CloudOpenAICloud scans of connected GitHub repos or new commits; Fix with Codex drafts a patch you review before it becomes a draft pull request; works in Codex desktop and web and requires Codex cloud30 September 2026Read
Cursor Rollouts and Security ReviewerCursorRollouts writes a monitoring plan for each pull request and rates staging and production after deploy; on a regression it can open a revert PR or hand the fix to a Cursor cloud agent; both bots are in the Automations tabTeams and Enterprise plans24 September 2026Read
Microsoft Copilot Code modeMicrosoftA Code mode for developers built on the same underlying tech as GitHub Copilot, with a Managed Copilot Runtime for code execution; rollout through the Frontier program25 September 2026Read
Replit AgentReplitThe CEO says Replit stopped working with Anthropic and announced an OpenAI partnership in August; a late September Replit blog post lists Claude Fable 5 and Claude Fable 5.1 in Replit Agent's production environment17 October 2026Read
Muse Code / Muse WonderMetaPer an unconfirmed Threads post: Muse Wonder is on the Meta Model API, and Muse Wonder Full goes only to Muse Code subscribers12 October 2026Read
DeepSeek Harness v0.2DeepSeekDesktop app for work and coding tasks, powered by DeepSeek models, with plugins and workspaces; users report it runs on macOS and Windows but not Linux2 October 2026Read

Milestones

  1. Cursor subagents get their own VMs.
  2. The Grok Build beta opens to SuperGrok and X Premium+ subscribers.
  3. Anthropic keeps Claude available in Cursor after SpaceX takes ownership.
  4. Meta’s Muse Code leaves beta with an SDK and subscriptions.
  5. GPT-6 Astra reaches Pro and Business before Plus.
  6. Claude Code starts reading AGENTS.md when a repo has no CLAUDE.md.
  7. JetBrains brings its agent tools together under one brand, Air.
  8. claude --cloud moves a Claude Code session to an Anthropic-managed VM that keeps working after the laptop closes.
  9. Cursor adds Rollouts, a bot that tracks each pull request into production, for Teams and Enterprise.
  10. Codex makes cloud environments reusable, so tasks keep running while your computer sleeps.
  11. OpenAI launches Ultrafast on GPT-6 Astra; in Codex it requires the new Pro 500 plan.
  12. OpenAI announces that Codex usage on the $200 Pro plan will fall from 20x to 10x of Plus.
  13. OpenAI enterprise customers can use GLM-5.3 Flash and Kimi K3 in Codex through Baseten.
  14. OpenAI’s Dots launch, and a user’s primary Dot is included in the plan.
  15. Day 1 of the 28-day Codex sprint: GPT-6 Astra and GPT-6.1 Sol run about 50% faster by default.
  16. Ultrafast mode reaches GPT-6.1 Sol in ChatGPT Work, Codex and the API.
  17. Claude Code Projects lets in its whole Pro and Max waitlist.
  18. GPT-6.1 Sol’s default context in Codex goes from 272k to 500,000 tokens.
  19. Cursor’s new plans include Grok usage and bill Claude at API list prices.
  20. The Claude Code desktop app gives each parallel task its own nested session.

FAQ

Does Cursor still include Claude?

Not on the new plans. The plans launched October 12 include Grok usage but no Claude. On the $30 Premium plan, Grok comes from an included weekly pool and every Claude request is billed on top at API list prices. The cheapest plan has no Claude at all. Current subscribers keep their included Claude usage until they switch (source).

What changes on the $200 ChatGPT Pro plan for Codex on October 30?

ChatGPT Work and Codex usage falls from 20x to 10x the Plus allowance, and GPT-6 Pro in Chat drops from 200 to 100 messages per week. Current limits hold through October 29. After that, existing subscribers get 62,500 usage credits, which expire December 31, 2026. The new $500 Pro plan gives 25x Plus and Ultrafast for GPT-6 Astra. The same post says Google’s and Anthropic’s $200 plans still give 20x (source).

Can Claude Code and Codex keep working after I close my laptop?

Both can. claude --cloud moves a Claude Code task to an Anthropic-managed VM that clones your branch from the GitHub remote. The VM has no separate compute charge, but it shares rate limits with the rest of your Claude usage (source).

Codex tasks keep running in the cloud while your computer sleeps, and each task starts from a published, reusable environment (source).

How much does Codex Ultrafast cost?

Ultrafast on GPT-6.1 Sol costs $12 per million input tokens and $60 per million output. Standard Sol stays at $2/$10. Astra’s Ultrafast in Codex required the Pro 500 plan, and OpenAI hasn’t said whether Sol follows the same rule (source).

Does Claude Code read AGENTS.md?

Yes, when the repo has no CLAUDE.md (source).

Latest