Claude Code vs Codex Pricing & Limits (November 2026): Which Plan, 5-Hour Windows, Weekly Caps

Which Plan Should You Buy?

Both tools start at $20/mo. The right plan depends on how much you use it, which model you run, and whether your work is long interactive sessions or fanned-out cloud tasks.

Codex Plus vs Claude Pro, Codex Pro vs Claude Max: Plan Matchups

Most searches compare one plan against its price twin. Here are the three price points side by side.

Claude Pro vs Codex Plus is the closest match. Both cap you every 5 hours. At $100 the plans split: Codex Pro has no 5-hour limit, while Claude Max keeps one at 5x Pro. Claude only drops the 5-hour limit at $200 with Ultra. For a team, Claude Team standard seats and ChatGPT Business seats both cost $20 per seat per month on annual billing.

What Changed in October and November 2026: Claude Code vs Codex Plans and Limits

Checked against claude.com/pricing and OpenAI's Codex pricing page on November 8, 2026.

The practical result: a 5-hour wall now only applies on the entry tiers. Codex removes it at $100. Claude removes it at $200. Weekly limits remain on every consumer tier of both tools.

What Changed in September 2026: Claude Code vs Codex Limits

Most Claude Code vs Codex limits pages were written before these dates. Each one changes the math.

The Claude math: if the original weekly cap was 100 units, summer users had 150 and now have 125. The Codex math: GPT-6 Astra allows 5-45 messages per window on Plus, under a third of GPT-6 Sol's 15-150. Switching the Codex model changes your effective plan size more than switching plans does.

How Each Tool Meters Usage

The unit of metering is the most important difference for picking a plan. They do not compare one to one.

The 5-Hour Window, Explained

Both tools reset a short-term allowance every 5 hours. What fills it is where they diverge.

Claude Code: a session budget, not a message count

Anthropic describes the Max plan as a session-based usage limit that resets every five hours, with Max at five times Pro's allowance. Ultra, the $200 tier, has no 5-hour limit for verified accounts and a weekly allowance of 10x Pro. It does not convert that to prompts. Anthropic says usage varies with conversation length, model choice, tool use, and effort level. An Opus 5.5 session with high effort and many tool calls drains the window faster than a short Sonnet session. That matters more now: since v2.1.280 (September 22, 2026) Claude Code starts Pro and Team Standard users on Opus instead of Sonnet. The 5-hour limits were not part of the weekly promotion that ended September 13, so they did not drop. On September 22 Anthropic raised them on Pro, Max, Team, and seat-based Enterprise with the Opus 5.5 launch. It did not publish the size of the increase.

Codex: messages per model, cloud included

Codex publishes a range per model. OpenAI's estimate for ChatGPT Plus is 5-45 GPT-6 Astra local messages per 5 hours, then GPT-6.1 Sol 15-160, GPT-6.1 Terra 35-285, GPT-6 Sol 15-150, GPT-6 Luna 350-3,000. Cloud chats share the same allowance, so a heavy cloud run eats into your local budget. Pro plans currently have no five-hour limit. GPT-5.4 and GPT-5.4 mini left Codex for ChatGPT sign-in on August 31, 2026, and GPT-5.5 followed on October 14.

Limit resets on both sides

Both vendors now hand out banked resets. With Opus 5.5, Anthropic gave subscribers a limit reset they can save. One reset refills the 5-hour or the weekly limit. You redeem it with "Reset for free" in Settings > Usage on the web or Claude Desktop, not from the Claude Code terminal, and it can expire. Claude Code v2.1.234 also added "Continue automatically at usage limit", which resumes work when the limit resets. Codex has banked rate-limit resets for Plus and Pro since June 2026, and its CLI rate-limit banner links to them.

Claude Code Weekly Caps by Plan

Claude Code adds a weekly limit on top of the 5-hour session limit. Anthropic publishes multipliers, not hours. Treat any page that quotes a fixed number of Sonnet or Opus hours per week as a third-party estimate.

Since September 14, 2026, weekly caps on Pro, Max, Team, and seat-based Enterprise are 25% above their pre-May baseline.

Codex Messages, Cloud Tasks, and Credit Rates

Codex meters messages per 5-hour window, then converts model usage into credits. Plus and Pro users who hit the limit can buy additional credits instead of upgrading.

Pro plans ($100, $200, $500) currently have no five-hour limit, so OpenAI prints no per-window range for them. Weekly limits may also apply. Enterprise and Edu on flexible pricing have no fixed rate limits. API-key users pay per token and get no cloud features.

Codex credit rates per 1M tokens

A typical GPT-5.6 Sol task uses 2-15 credits. Fast mode uses included usage at 2.5x the Standard rate and purchased credits at 2x; Ultrafast uses 8x and 6x. GPT-6 Astra output costs 5x GPT-6 Sol per token, which is why its message range is under a third of Sol's. Credit rates are 25x the API dollar price, with no cache-write charge. Source: OpenAI Codex pricing.

Plan-by-Plan Pricing

The ChatGPT Pro $100 tier launched April 9, 2026. OpenAI paused new $200 sign-ups on September 10 and now sells Pro at $100, $200, and $500. Claude Max and Ultra also include monthly API credits ($100 and $200) for the Claude Platform. See Morph pricing if you want to run open-source models per token instead of paying for a subscription window.

Install and Sign In

Claude Code

Native install (recommended). macOS, Linux, and WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

Also available via brew install --cask claude-code (stable channel) or claude-code@latest, winget install Anthropic.ClaudeCode, signed apt/dnf/apk repos, and npm install -g @anthropic-ai/claude-code. The npm package needs Node.js 22 or later as of v2.1.198. Native installs auto-update; set "autoUpdatesChannel": "stable" to stay about a week behind latest. Requirements: macOS 13.0+, Windows 10 1809+, Ubuntu 20.04+, Debian 10+, Alpine 3.19+, 4 GB+ RAM. The current release is v2.1.319 (November 6, 2026). Since v2.1.280 (September 22, 2026), Claude Code defaults to Claude Opus 5.5 on every subscription plan.

OpenAI Codex

Mac and Linux:

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Windows:

powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

Also npm install -g @openai/codex or brew install --cask codex. Run codex and choose "Sign in with ChatGPT." Surfaces: the CLI, IDE extensions for VS Code, Cursor, and Windsurf, a desktop app (codex app; since July 9, 2026 Codex lives inside the ChatGPT desktop app), and Codex Web at chatgpt.com/codex. Codex CLI 0.156.0 (September 22, 2026) turns /usage into an analytics dashboard with account usage, token totals, and plugin and skill activity. It also adds an optional fullscreen /tui, turns voice conversations on by default (F8 toggles), and enables worktree support by default. The current stable release is 0.167.0 (November 6, 2026).

MCP support

Both support Model Context Protocol servers. Claude Code v2.1.280 added CLAUDE_CODE_MAX_MCP_DESCRIPTION_LENGTH, which caps each MCP tool description (default 2,048 characters). Lowering it trims the tool schema sent on every turn, which saves context and, on a subscription, some of your window.

If You Run Models Yourself Instead

Subscription caps exist because frontier models are expensive to serve. If your workflow is API-driven, or you want open-source models with no 5-hour window or weekly cap, the numbers that matter are per-token cost and output fidelity.

Morph serves DeepSeek with 16-bit (bf16) activations and no fp8/int8 activation quantization. Many serverless providers quantize activations to fp8 to cut cost, which shifts outputs away from the reference weights. Morph also runs codegen-specific speculative decoding and custom inference kernels tuned for coding agents.

See Morph models and pricing for the full list.

Decision Framework

Frequently Asked Questions

Source for the Hacker News figure: joshstrange on HN, September 22, 2026. For benchmarks and architecture, read the Codex vs Claude Code comparison.