Blog

Codex pricing in November 2026: every ChatGPT plan, the API-key rates, and what a month of use costs

Ramashankar S · · Updated · 17 min read

Codex pricing comes in two forms. Codex is included in every ChatGPT plan: Free ($0) and Go ($8) get GPT-6 Luna in the desktop app, Plus costs $20 a month, Pro costs $100, $200 or $500, and Business seats cost $25 or $125 a month ($20 or $100 billed yearly). Or you sign in with an OpenAI API key and pay per token: GPT-6.1 Sol, the model OpenAI points paid plans to, costs $2.00 / $10.00 per million tokens, with cached input at $0.10.

At API rates a typical Codex message costs about $0.19, so a light month of 60 messages comes to $11.52, daily use (600 messages) to $115.20 and heavy use (2,400) to $460.80. Plus starts to cost less than the API at about 104 messages a month, if its allowance covers your use. Every price below was checked on OpenAI's pages on 5 November 2026, against Codex CLI 0.166.0, the current release (29 October).

Codex pricing by plan

There is no separate Codex subscription. Codex and ChatGPT Work share one allowance on every plan, and the Codex pricing page lists what each plan gets:

PlanPrice a monthWhat Codex getsCloud tasks and GitHub review5-hour window
Free$0GPT-6 Luna in the desktop app, rolling outNoNot published
Go$8GPT-6 Luna in the desktop app, rolling outNoNot published
Plus$20CLI, IDE extension, web, iOS; GPT-6.1 Sol, Luna and limited AstraYes15 to 160 GPT-6.1 Sol messages
Pro 100$100More usage, full Astra allowanceYesNone, weekly limit applies
Pro 200$200More usage than Pro 100YesNone, weekly limit applies
Pro 500$500Highest usage, Ultrafast modeYesNone, weekly limit applies
Business Standard$25, or $20 billed yearlyPlus-level usage, admin controls, no training on your dataYes15 to 160 GPT-6.1 Sol messages
Business Premium$125, or $100 billed yearlyHigher usage, full Astra allowanceYesNone
Enterprise, EduCustomShared pool, or USD per tokenYesSet by the agreement
API keyPer tokenAny model your API organization can useNoNone, API rate limits only
Codex pricing by plan: monthly price, models, cloud access, 5-hour window and what happens past the limit, from Free to API key

Business needs at least two seats, and you can mix Standard and Premium, per OpenAI's Business overview. Pro is billed monthly only. OpenAI's Pro tiers page, updated 3 November, says Pro 100 and Pro 500 now get twice their previous Codex and Work allowance, and every Pro 200 subscriber now has the higher of its two former allowances. Plus didn't change.

One change is already scheduled: "Starting in December", Plus, Pro and Business will share one token-based allowance across ChatGPT and Codex, per OpenAI's usage-limits article. Instant chat won't count against it. Until then, regular ChatGPT chats have their own limits, while Work already shares Codex's.

Codex usage limits on each plan

Plus and Business Standard get an estimated 15 to 160 GPT-6.1 Sol messages per 5 hours, plus a weekly limit OpenAI doesn't publish. Pro and Business Premium have no 5-hour window. A message here is one prompt you send, with every model call and tool step it triggers:

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

OpenAI's Business limits page gives 15 to 150 for GPT-6.1 Sol on Standard seats, against 15 to 160 on the pricing page. OpenAI puts the spread down to task size: a small script uses a fraction of what a long task over a large project does.

Enterprise and Edu workspaces on flexible pricing have no fixed limits and draw from a shared pool; without flexible pricing, seats get Plus-level limits. In the CLI, /status shows both your 5-hour and weekly meters, and the usage dashboard shows reset times.

Some settings drain the allowance faster. Fast mode uses it at 2.5 times the Standard rate, Ultrafast on Pro 500 at 8 times, and image generation at 3 to 5 times a similar turn. Voice in the desktop app costs $0.05 a minute of usage.

What extra usage costs

At the limit, a turn already running can finish. After that, Plus and Pro users can buy extra usage, Business owners buy a shared workspace pool, and anyone can switch to an API key. OpenAI sells that extra usage as ChatGPT credits, and its per-token rate card is the API price list in a different unit: GPT-6.1 Sol costs 50 units per million input tokens, and OpenAI's Enterprise USD rate card lists the same model at $2.00, so one unit is worth $0.04. OpenAI's student offer confirms it, valuing 2,500 units at $100.

One difference makes extra usage on a plan slightly less expensive than an API key: Codex's rate card has no cache-write charge. Purchased usage lasts 12 months, is generally non-refundable, and automatic reload can be capped per month, according to OpenAI's personal-plan FAQ.

Cloud tasks vs local CLI vs IDE extension: what counts

The CLI, the IDE extension and local chats in the desktop app all draw from the same allowance as "local messages". Codex Cloud tasks draw from it too, and OpenAI warns they "may use more of your allowance than local messages". What changes with an API key is which surfaces exist at all:

SurfaceSigned in with ChatGPTSigned in with an API key
CLI, IDE extension, desktop appPlan allowancePer token at API prices
codex exec and the Codex SDKPlan allowancePer token at API prices
Codex Cloud tasksPlan allowance, Plus and above onlyNot available
GitHub review (@codex review, automatic reviews)Code Review usage, runs on GPT-5.3-CodexNot available
/review in the CLIPlan allowancePer token at API prices
Slack and LinearPlan allowanceNot available
Fast mode2.5x allowance2x the token price

OpenAI's Codex help article adds two details: standard Codex Cloud environments have no separate VM charge "at launch", and auto-review safety checks are free when you sign in with ChatGPT. Codex Cloud isn't included on Free or Go.

Meet the all new Codex CloudOpenAI · 3:24 · YouTube
OpenAI's walkthrough of cloud environments and tasks, the part of Codex that needs a ChatGPT plan.

Codex with an API key: per-token prices

With an API key, Codex bills each request at the OpenAI API price list, and you can use any model your API organization has. Standard prices per million tokens, for requests up to 272,000 input tokens:

ModelInputCached inputCache write (30 min)OutputNote
gpt-6.1-sol$2.00$0.10$2.50$10.00OpenAI's pick for paid plans
gpt-6-astra$10.00$1.00$12.50$50.00Hardest work
gpt-6.1-terra$1.00$0.05$1.25$5.00Needs CLI 0.159.0 or later
gpt-6-sol$2.00$0.20$2.50$10.00Double 6.1 Sol's cached price
gpt-6-luna$0.10$0.01$0.125$0.50Focused, repeatable tasks
gpt-5.3-codex$1.75$0.175none listed$14.00Shuts down 1 Apr 2027
gpt-5.5$5.00$0.50none listed$30.00Left ChatGPT sign-in 14 Oct

GPT-5.3-Codex is the only Codex-branded model left on the API, and OpenAI's deprecations page shuts it down on 1 April 2027, naming gpt-6-sol as the replacement. gpt-5-codex, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini and gpt-5.2-codex shut down on 23 July 2026. GPT-5.5 left Codex for ChatGPT sign-in on 14 October but still works with an API key.

Three API rules raise the bill compared with a plan:

  • On GPT-5.6 and later, new input written to the prompt cache costs 1.25 times the input price, per OpenAI's prompt caching guide. A plan charges no cache writes.
  • Above 272,000 input tokens, a request costs 2 times input and 1.5 times output: $4.00 / $15.00 on GPT-6.1 Sol. Subscription usage of GPT-6.1 Sol in Codex has no such surcharge.
  • Fast mode doubles the token price: $4.00 / $20.00 on GPT-6.1 Sol.

Pro and Business plans include a monthly grant of free API usage ($100, $200 or $500 on Pro), but OpenAI's benefit page says "Codex usage through API keys is not supported by this benefit." It pays for your own apps, not for Codex.

To switch Codex to a key, pipe it to the login command, then check which method is active. The Codex CLI install guide covers setup, and the OpenAI token pricing reference has every other model's rates.

printenv OPENAI_API_KEY | codex login --with-api-key
codex login status

What Codex costs per month: light, daily and heavy use

On an API key with GPT-6.1 Sol, Codex costs about $12 a month for light use, $115 for daily use and $461 for heavy use. Those figures assume one GPT-6.1 Sol message, at default effort, uses 600,000 input tokens across its model calls, 95% of them read from the cache, and 6,000 output tokens including reasoning. Agent loops resend the same context on every call, which is why the cache share is high: the sample turn in OpenAI's non-interactive mode docs read 24,448 of its 24,763 input tokens from cache.

LineTokensAPI priceAPI costPlan rate (no cache write)
Cached input570,000$0.10 per 1M$0.057$0.057
New input30,000$2.50 per 1M (cache write)$0.075$0.060
Output6,000$10.00 per 1M$0.060$0.060
One message$0.192$0.177

That lands close to OpenAI's own figures. Its Business rate card puts a typical GPT-6.1 Sol task at 2 to 15 units ($0.08 to $0.60), and its legacy per-message table averages about 4 units ($0.16) per GPT-6.1 Sol local message. It also says Codex costs about $100 to $200 per developer per month on average, which brackets the daily profile below.

Three profiles, with the same message size on every model:

ModelPer messageLight: 60 a monthDaily: 600 a monthHeavy: 2,400 a month
GPT-6 Luna$0.012$0.75$7.47$29.88
GPT-6.1 Terra$0.096$5.76$57.60$230.40
GPT-6.1 Sol$0.192$11.52$115.20$460.80
GPT-5.3-Codex$0.236$14.17$141.75$567.00
GPT-6 Sol$0.249$14.94$149.40$597.60
GPT-6 Astra$1.245$74.70$747.00$2,988.00
What Codex costs a month on an API key for light, daily and heavy use on GPT-6 Luna, GPT-6.1 Terra, GPT-6.1 Sol and GPT-6 Astra

Light is a few evenings a week, daily is 30 messages a working day, and heavy is 120 a day, often with parallel agents. Real token counts differ by model: on the official Terminal-Bench 4.0 leaderboard, GPT-6 Sol in Codex used 5.0 billion tokens to GPT-6.1 Sol's 1.5 billion on the same tasks, so their real gap is wider. Long sessions push the other way: if every call in a message sat above 272,000 tokens, the GPT-6.1 Sol message would cost $0.354, 1.84 times as much. For the general formula, see how much an LLM API costs per month.

When an API key costs less than a plan, and when it doesn't

An API key costs less than Plus below about 104 GPT-6.1 Sol messages a month, and less than Pro 100 below about 521. Above those points the plan costs less, if its allowance covers your use. Each point is the plan price divided by $0.192:

PlanPrice a monthAPI key costs less below
Plus$20104 messages
Business Standard$25130 messages
Pro 100$100521 messages
Business Premium$125651 messages
Pro 200$2001,042 messages
Pro 500$5002,604 messages
Break-even between Codex plans and an API key, in GPT-6.1 Sol messages a month, with the light, daily and heavy profiles marked

OpenAI says API prices shouldn't be used to estimate how many tasks a plan includes, so read this as a money comparison only. A plan comes out ahead only if its allowance actually covers your messages.

An API key costs less when:

  • You send fewer than about 100 GPT-6.1 Sol messages a month and don't need ChatGPT itself. The light profile costs $11.52.
  • Most of your work runs on GPT-6 Luna. Even the daily profile costs $7.47.
  • Codex runs in CI or scripts. OpenAI recommends an API key for automation, and an API key has no 5-hour window, only your organization's rate limits.

A plan costs less when:

  • You send more than about 104 GPT-6.1 Sol messages a month. The daily profile costs $115.20 on a key.
  • Your sessions run long. Plans charge no long-context surcharge on GPT-6.1 Sol; the API bills those requests at up to 1.84 times as much in this model.
  • You use GPT-6 Astra. At $1.245 a message, 100 Astra messages cost more on a key than Pro 100 does.
  • You need Codex Cloud, GitHub review or Slack, which an API key can't use.

Between Plus and Pro 100 there's a third route: stay on Plus and buy extra usage at the limit. Pro 100 costs $80 more than Plus, and $80 of extra usage buys about 452 GPT-6.1 Sol messages at the plan rate of $0.177. If your overflow beyond what Plus covers stays under about 450 messages a month, Plus with extra usage costs less than Pro 100.

Price your own Codex runs

codex exec --json prints a turn.completed event with the turn's token usage, whether you're signed in with ChatGPT or a key. This script prices it at API rates and at the plan rate. In Codex 0.166.0, output_tokens already includes reasoning, and input_tokens includes the cached and cache-write tokens.

# price_codex_run.py: price a `codex exec --json` run at OpenAI's list prices
import json
import sys

# US$ per 1M tokens, checked 5 Nov 2026: input, cached input,
# 30-minute cache write, output (requests under 272K input tokens)
PRICES = {
    "gpt-6.1-sol": (2.00, 0.10, 2.50, 10.00),
    "gpt-6.1-terra": (1.00, 0.05, 1.25, 5.00),
    "gpt-6-luna": (0.10, 0.01, 0.125, 0.50),
    "gpt-6-astra": (10.00, 1.00, 12.50, 50.00),
}

p_in, p_cached, p_write, p_out = PRICES[sys.argv[1]]
used = {"input_tokens": 0, "cached_input_tokens": 0,
        "cache_write_input_tokens": 0, "output_tokens": 0}  # output includes reasoning

for line in sys.stdin:
    if not line.startswith("{"):
        continue
    event = json.loads(line)
    if event.get("type") == "turn.completed":
        for key in used:
            used[key] += event["usage"].get(key, 0)

cached, written = used["cached_input_tokens"], used["cache_write_input_tokens"]
plain = used["input_tokens"] - cached - written
api_usd = (plain * p_in + cached * p_cached + written * p_write
           + used["output_tokens"] * p_out) / 1e6
# ChatGPT credit rates are 25x these dollar prices, with no cache-write charge
plan_usd = ((plain + written) * p_in + cached * p_cached
            + used["output_tokens"] * p_out) / 1e6

print(used)
print(f"API key: ${api_usd:.4f}   plan usage at $0.04 a credit: ${plan_usd:.4f} ({plan_usd / 0.04:.2f} credits)")
codex exec --json -m gpt-6.1-sol "fix the failing test in tests/test_dates.py" \
  | python3 price_codex_run.py gpt-6.1-sol

Fed the 600,000-token message from the model above, it prints API key: $0.1920 plan usage at $0.04 a credit: $0.1770 (4.42 credits). Run it on a week of real tasks before choosing a plan.

If you use a key, give Codex its own API project with a hard spend limit. When the project reaches it, requests fail with a 429 project_spend_limit_exceeded, though OpenAI's spend-limits guide notes enforcement isn't instant. In CI, set CODEX_API_KEY only on the Codex step, never as a job-wide variable. The OpenAI spending-limit guide walks through the settings.

What Codex users report about limits and cost

  • GPT-6.1 Sol in Codex scored 58.2% on Terminal-Bench 4.0 for about $600 in total, the same score GPT-6 Astra reached for about $3,300 (official Terminal-Bench leaderboard, checked 5 Nov 2026).
  • OpenAI's own planning figure is about $100 to $200 per developer per month for Codex, varying with model, concurrent instances, automations and Fast mode (OpenAI Business rate card, checked 5 Nov 2026).
  • "Their 5 hour limit works better with a 'constant background trickle' usage pattern." (a Hacker News commenter, 1 Nov 2026, link)
  • "no indication at all what amount of usage you get in a subscription" (a Hacker News commenter on the December merge, 2 Nov 2026, link)
  • During a read-only audit with two subagents on Luna at xhigh effort, a user's 5-hour meter went from 22% to 98% in about 117 seconds while about 4.4 million input tokens were processed (GitHub issue on openai/codex, 2 Nov 2026, link).

Where Aight fits, and where it doesn't

Aight can't run Codex. Codex speaks OpenAI's Responses API, and OpenAI's Codex model docs say current releases don't support a Chat Completions-only endpoint. Aight documents only Chat Completions, embeddings and the model list, so for Codex itself the choice is a ChatGPT plan or an OpenAI API key, as above.

Aight is useful for the scripted work around a coding agent: a CI step that summarizes a diff, sorts a failing test log or drafts release notes through Chat Completions. Aight lists gpt-6.1-sol at OpenAI's $2.00 / $10.00 with $0.10 cached input, next to Claude, Gemini and DeepSeek models, under one key. Each key can carry a monthly cap that returns 403 key_cap_reached, a list of allowed models and an expiry date. Every recharge adds 10% extra balance, about $0.94 per $1 of list-price usage after the 3% processing fee outside India. OpenAI says GPT-6.1 Sol needs the Responses API for tool calling, so keep Sol on Aight to plain-text jobs like these, and check each model's maximum output on the models page. The Aight docs show the configuration.

FAQ

Is Codex free?

Yes, in a limited form. ChatGPT Free and Go include Codex with GPT-6 Luna in the desktop app, still rolling out, with no Codex Cloud. For the CLI with GPT-6.1 Sol, cloud tasks and GitHub review, you need Plus at $20 a month or higher.

Is Codex included in ChatGPT Plus?

Yes. Plus includes Codex in the CLI, IDE extension, web and iOS, with GPT-6.1 Sol, Luna and limited Astra, plus Codex Cloud and GitHub review. OpenAI estimates 15 to 160 GPT-6.1 Sol messages per 5 hours, with a weekly cap it doesn't publish.

Do Pro's free API dollars pay for Codex?

No. OpenAI's benefit page says Codex usage through API keys isn't covered, and the grant doesn't raise your Codex limits either. It covers your own apps on the Responses, Agents and other APIs.

What happens when you hit the Codex usage limit?

The turn in progress can finish. Then you wait for the 5-hour or weekly reset shown by /status, buy extra usage on Plus or Pro, upgrade, or sign in with an API key for local work at standard API prices. OpenAI support can't reset limits.

Does Codex cost less than Claude Code?

On Terminal-Bench 4.0 at API prices, the GPT-6.1 Sol run in Codex cost about $600 against $4,700 for Claude Opus 5.5 in Claude Code, which scored 64.8% to Codex's 58.2%. Subscriptions are a different comparison; see Claude Code pricing and Codex vs Claude Code.


Sources

Checked on 5 November 2026.