Codex usage limits depend on your ChatGPT plan, the model you pick and how big each task is, not on a fixed message count. On Plus and Standard Business, OpenAI estimates 15 to 160 local messages per five hour window with GPT-6.1 Sol, and weekly limits can apply on top; Pro plans currently have no five hour limit. The big change is coming in December 2026: Plus, Pro and Business plans will draw ChatGPT and Codex from one shared allowance measured in tokens, so a heavy Codex day can also cut into your ChatGPT chats (Instant chat stays free of the meter). You can see what is left with /status in the Codex CLI or in Settings, Usage.

This guide covers how the limits work today, what the shared allowance changes, how credits and resets work, and the settings that make an allowance last longer. Everything here comes from OpenAI's Codex pricing page, its help article on ChatGPT and Codex usage limits, the Pro tiers article, the credits article and the ChatGPT release notes, all read on November 2, 2026. The Codex pricing page, the GPT-6.2 launch post and the ChatGPT Work and Codex help article were rechecked on November 19, 2026.

What's new (November 19, 2026): GPT-6.2 Astra and GPT-6.2 Sol arrived in Codex on November 17, in the Codex app and CLI and in ChatGPT Work. OpenAI's launch post lists them for Plus, Pro, Enterprise and Edu users. The launch post does not name Business. The Codex pricing page, checked on November 19, now says both models are available on Plus, Pro, Business, Enterprise and Edu in ChatGPT Work and in the Codex app and CLI, and that they are off by default in Enterprise and Edu workspaces until an administrator enables them. The same page estimates 6 to 50 local messages per five hours with GPT-6.2 Astra and 17 to 180 with GPT-6.2 Sol on Plus and Standard Business, a little more than GPT-6 Astra and GPT-6.1 Sol. GPT-6.2 Astra costs 200 credits per million input tokens, down from 250 for GPT-6 Astra, and GPT-6.2 Sol keeps GPT-6.1 Sol's credit rates. Ultrafast now reaches Pro 200 too, starting with GPT-6.2 Sol. The ChatGPT Work and Codex help article says the models you see depend on your plan, seat and workspace settings, and that Ultrafast stays unavailable on Plus, Pro $100 and Business.

What changes in December: one allowance for ChatGPT and Codex

Until now, ChatGPT chat and OpenAI Codex were metered separately on personal plans. Codex already shared its allowance with ChatGPT Work (the agent mode inside ChatGPT) and with ChatGPT for Excel, Word and PowerPoint, but ordinary chats did not touch it. OpenAI's help article says that starting in December, the usage allowance on Plus, Pro and Business plans will be shared across ChatGPT and Codex, as Work's agent features move into the main ChatGPT experience.

Four details matter:

  • Usage is measured in tokens. How much you burn depends on the model and the task; bigger models and complex jobs use more.
  • Instant chat does not count. OpenAI says Instant chat will not draw from your usage, so quick questions stay unmetered.
  • One place to check. Combined remaining usage and reset times will show in Settings, Usage.
  • Same three ways out. When you hit the limit, you wait for the reset, upgrade, or buy credits.

OpenAI says it expects most users won't need to wait for a reset more often than they do now. The worry, voiced in the Hacker News thread about the change (227 points), is the reverse case: a developer who spends the whole allowance in Codex will then find reasoning chats in ChatGPT blocked until the reset. OpenAI has not published a reserved slice for chat, and no exact December date is given.

How Codex usage limits work today

The pricing page is clear that its numbers are estimates, not quotas. Model choice, context size, reasoning, tool calls, retrieval and caching all change what a message costs, so two similar looking tasks can use very different amounts.

These are OpenAI's estimates of local messages per five hour window on Plus and Standard Business:

ModelLocal messages per 5 hours
GPT-6.2 Astra6 to 50
GPT-6.2 Sol17 to 180
GPT-6 Astra5 to 45
GPT-6.1 Sol15 to 160
GPT-6.1 Terra35 to 285
GPT-6 Sol15 to 150
GPT-6 Luna350 to 3,000

A few rules sit around that table:

  • Weekly limits may also apply, on top of the five hour window.
  • Cloud tasks cost more. Local messages and cloud chats share the allowance, and cloud tasks may use more of it.
  • Pro currently has no five hour limit, though weekly limits may apply.
  • Enterprise and Edu with flexible pricing have no fixed rate limits; usage draws on credits. Without flexible pricing, those seats get the same per seat limits as Plus for most features.
  • Speed modes cost extra. Fast mode uses your included allowance at 2.5 times the Standard rate on older models and 2 times on GPT-6.2 Astra and Sol. Ultrafast uses it at 8 times on GPT-6 Astra and GPT-6.1 Sol and 6 times on GPT-6.2 Astra and Sol. Among personal plans, Pro 500 gets Ultrafast on both GPT-6.2 models, and Pro 200 is getting it, starting with GPT-6.2 Sol.
  • Image generation uses included limits about 3 to 5 times faster than a similar turn without images.

If you hit the limit in the middle of a turn, OpenAI says the agent can finish that turn, subject to fair use limits.

Codex limits on Plus vs Pro vs Business

PlanPriceCodex limits
Free$0GPT-6 Luna in the desktop app, subject to rollout
Go$8 a monthGPT-6 Luna in the desktop app, subject to rollout
Plus$20 a monthFive hour window plus weekly limits; web, CLI, IDE extension, iOS and cloud integrations
Pro 100, 200, 500$100, $200, $500 a monthNo five hour limit; weekly allowance grows by tier; Ultrafast on Pro 500, and on Pro 200 starting with GPT-6.2 Sol
Business$25 per user a month billed monthlyStandard seats match the Plus estimates; workspace credits
API keyAPI ratesNo subscription limit; no cloud features such as GitHub review or Slack

The Pro tiers changed twice this autumn. On October 29, OpenAI doubled the Work and Codex allowance on Pro 100 and Pro 500, canceled the Pro 200 reduction it had announced on September 29, and raised Pro 200 subscribers on the lower allowance to the same level. Plus usage did not change. OpenAI does not publish the size of any allowance in tokens, and the Pro article warns that a larger allowance does not guarantee twice as many finished tasks.

For how Codex compares with other agents on price and features, see Claude Code vs Codex vs Gemini CLI and our best AI coding assistant roundup.

How to check your Codex usage and reset time

  • In the CLI: type /status during a session to see your remaining limits.
  • In ChatGPT: Settings, Usage shows your balance today and will show the combined allowance and reset times after the December change.
  • In the Codex app: Usage and Billing shows your credit balance and lets you buy more.

OpenAI suggests checking the dashboard every week or two to learn your pace. Support cannot reset limits for you; per the Pro article, it can only investigate if usage was counted wrongly or access did not come back after a reset. Some Plus and Pro accounts also get banked rate limit resets through Codex invitations, which the June release notes introduced; the invitation dialog shows the current reward and when it expires.

Codex credits: what they cost in practice

Credits are the pay as you go layer. Your plan's included usage goes first, then usage draws from your credit balance. Credits work across Codex, ChatGPT Work, Word, Excel and PowerPoint, last 12 months from purchase, are not refundable except in narrow cases, and are separate from OpenAI API credits. Prices for a credit pack vary by plan and region and are shown at checkout. If you turn on automatic reload, set the maximum monthly spend so a runaway agent cannot keep buying.

The rate card lists credits per million tokens at Standard speed:

ModelInputCached inputOutput
GPT-6.2 Astra200101,000
GPT-6 Astra250251,250
GPT-6.2 Sol502.5250
GPT-6.1 Sol502.5250
GPT-6.1 Terra251.25125
GPT-6 Luna2.50.2512.5

On purchased credits, Fast mode costs 2 times these rates and Ultrafast 6 times (different from the multipliers on included usage). GPT-6.2 Astra and Sol Ultrafast costs 3 times on purchased credits through February 28, 2027, then 6 times from March 1. Agent sessions reread the same context over and over, so cached input is often most of the volume, which is why the cached column matters so much. This short script estimates a session's credit cost from the token counts Codex reports:

RATES = {  # credits per 1M tokens: input, cached input, output (Standard speed)
    "gpt-6-astra":   (250, 25, 1250),
    "gpt-6.1-sol":   (50, 2.5, 250),
    "gpt-6.1-terra": (25, 1.25, 125),
    "gpt-6-luna":    (2.5, 0.25, 12.5),
}
SPEED = {"standard": 1, "fast": 2, "ultrafast": 6}  # purchased credit multipliers

def credits(model, input_tokens, cached_tokens, output_tokens, speed="standard"):
    inp, cached, out = RATES[model]
    uncached = input_tokens - cached_tokens
    base = (uncached * inp + cached_tokens * cached + output_tokens * out) / 1_000_000
    return round(base * SPEED[speed], 2)

print(credits("gpt-6.1-sol", 2_000_000, 1_600_000, 60_000))  # 39.0

We ran it for a session that read 2 million input tokens, 1.6 million of them cached, and wrote 60,000. Results: GPT-6 Astra 215 credits, GPT-6.1 Sol 39, GPT-6.1 Terra 19.5, GPT-6 Luna 2.15. The same Sol session in Fast mode came to 78, and with no cache hits at all it jumped to 115. Model choice and cache reuse move the bill far more than prompt length.

How to make your Codex allowance last longer

OpenAI's own tips, with what they mean in practice:

  1. Pick the smallest model that does the job. Terra and Luna estimates run well above Sol's per window. Save Astra for the hard problems.
  2. Use Fast mode on purpose, not by default. At 2.5 times the included rate it drains a window quickly.
  3. Trim AGENTS.md. Every line is sent with every task. On a large repo, nest smaller AGENTS.md files in subfolders so only the relevant ones load; our AGENTS.md guide shows the layout.
  4. Turn off MCP servers you are not using. Each server's tool list adds context to every message. Code Mode for MCP explains why big tool lists are expensive.
  5. Scope the task. Point Codex at the files that matter, define the output and keep optional polish separate.
  6. Run local tasks rather than cloud ones when you are near a limit, since cloud tasks can cost more.

When the allowance runs out, there are two routes besides waiting or buying credits. The pricing page says anyone can run extra local chats with an API key at standard API rates (codex login --with-api-key reads the key from stdin). The CLI also has an --oss flag (with --local-provider set to lmstudio or ollama) that runs Codex against a local model instead of OpenAI's, which suits small edits on a machine with enough memory. Other agents are an option too: see Claude Code background agents for running several sessions on your own machine, or best open source coding agents for agents that work with any model.

FAQ

When do Codex usage limits reset?

There are two clocks on Plus and Standard Business: a five hour window and, where it applies, a weekly limit. Pro currently has no five hour window. The exact reset time for your account shows in /status in the CLI and in Settings, Usage. OpenAI does not publish fixed reset times.

How many Codex messages do I get on ChatGPT Plus?

OpenAI estimates 17 to 180 local messages per five hour window with GPT-6.2 Sol, 15 to 160 with GPT-6.1 Sol, 35 to 285 with GPT-6.1 Terra, 6 to 50 with GPT-6.2 Astra, 5 to 45 with GPT-6 Astra and 350 to 3,000 with GPT-6 Luna. These are ranges, not quotas, because large tasks use far more per message.

Will ChatGPT chats use my Codex limits?

Starting in December 2026, yes on Plus, Pro and Business: ChatGPT and Codex will draw from one token based allowance. Instant chat is the exception and will not draw from it. Before December, ordinary chats are metered separately.

Do Codex credits expire?

Yes. Credits are valid for 12 months from purchase and do not roll over. They are also not API credits, so they cannot pay for direct API calls.

Can I use Codex without hitting ChatGPT limits?

Yes, by signing in with an API key. Usage is then billed at API rates with no subscription limit, but you lose cloud features such as GitHub code review and the Slack integration.