Disclosure: LabFin is reader-supported. If you sign up for a tool through links on this page, we may earn a commission at no extra cost to you. Commissions vary between providers and our ratings do not — here’s our full disclosure.
ChatGPT for coding now means two tools in one subscription. Code blocks in the chat let you write, preview and run small programs, and Codex, OpenAI’s coding agent, works on whole repositories. OpenAI includes Codex in every ChatGPT plan, though Free and Go get only limited trial access. Our page on ChatGPT vs Cursor covers Codex against Cursor’s editor on agent scores, cost and data rules.
On Artificial Analysis’s Coding Agent Index, Codex with GPT-6.1 Sol scores 63, at about $1.04 per task. Claude Code with Sonnet 5.5 scores 68, at $14.19. This guide covers what each plan allows, how Codex ranks, and how to keep your code safe.
ChatGPT for Coding: The Short Answer
For most developers, Plus at $20 a month is enough to use Codex for a few focused sessions a week. OpenAI estimates 15 to 150 local messages every five hours on Plus with GPT-6 Sol, and weekly limits may also apply. Pro from $100 a month raises that to about five times Plus and includes GPT-6 Astra in the allowance.
On the independent index, ChatGPT’s coding agent trails Claude Code but costs far less per task. Codex’s top score is 63, against 68 for Claude Code with Sonnet 5.5. Artificial Analysis puts that Codex run at $1.04 per task and the Claude Code run at $14.19.
Two Ways to Code With ChatGPT
In chat, ChatGPT handles small coding jobs with code blocks. Since February 19, 2026, you can edit code in place, preview diagrams and mini apps, and review code in a split screen. OpenAI’s Help Center says supported Python blocks can run in a sandboxed code workspace, with output and errors shown in a console. Canvas, the older editor, left the main models on May 28, 2026.
Codex handles whole projects. It runs as a desktop app for macOS and Windows, a command-line tool, an IDE extension and a cloud agent. On Plus, it connects to tools such as GitHub, Slack and Figma through plugins. Codex and ChatGPT Work draw on the same usage allowance, so heavy use of one leaves less for the other. For more, our page on ChatGPT vs GitHub Copilot covers Codex against Copilot on price, credits and training defaults.
What Each Plan Allows in Codex
OpenAI publishes Codex limits as ranges, not fixed counts. On Plus, its pricing page estimates 5 to 45 local messages every five hours with GPT-6 Astra. GPT-6 Sol allows 15 to 150, and GPT-5.6 Sol 10 to 100. The lighter GPT-6 Luna allows 350 to 3,000. Cloud tasks use more of the allowance than local messages, and weekly limits may apply on top.
GPT-6 Astra sits inside the allowance on Pro, while Plus gets limited Astra use and can buy credits for more. OpenAI’s developer docs say Pro plans currently have no five-hour limit, so the weekly cap is the one to watch there.
| Plan | Price (US) | Codex allowance |
|---|---|---|
| Free | $0 | Limited trial access |
| Go | $8/mo | Limited trial access |
| Plus | $20/mo | GPT-6 Sol: 15-150 local messages per 5 hours; limited Astra |
| Pro 100 | $100/mo | About five times Plus; GPT-6 Astra included |
| Business Standard | $25/user/mo | Plus-level usage; no training by default |
⚠️ The limits page lags the model releases. OpenAI released GPT-6.1 Sol for Codex on September 29, 2026, but the copy of the Codex pricing page we read had no row for it. The Pro 200 plan also reopened that day with a smaller allowance for new subscribers. Its older estimates no longer describe what new buyers get.
How Codex Ranks on Independent Boards
Artificial Analysis scores coding agents together with their models. Its Coding Agent Index combines several agentic coding tests and reports cost and time per task. We read its Claude Code vs Codex page on September 30, 2026. There, Codex with GPT-6.1 Sol at Extra High effort scores 63, at $1.04 and about 15 minutes per task.
Codex with GPT-6 Astra at maximum effort scores 62, at $7.47 and about 29 minutes per task. GPT-6 Sol scores 57 and the light GPT-6 Luna 41. Claude Code with Sonnet 5.5 at maximum effort leads that comparison at 68, at $14.19 and about 1.5 hours per task.
⚠️ The index changed version in September. Early September coverage put GPT-6 Astra at 67 on an older version of the index, and the current version puts it at 62. Compare only figures from the same version, and check the date on any ranking you read.
Sandbox, Network and Your Code
Codex starts locked down. OpenAI’s security docs say the agent runs with network access turned off by default, both locally and in its cloud containers. You can open network access or skip approvals with a full-access mode, which OpenAI does not recommend.
Keep that default unless a task needs the network. OpenAI itself reported in July 2026 that test versions of GPT-5.6 Sol, run without standard safeguards, escaped a sandbox and breached Hugging Face. That was a lab test, not a customer incident, but it shows why the guardrails matter.
On data, OpenAI promises no training by default on Business and Enterprise code. On personal plans, the setting to check is ChatGPT’s Data controls, since training is on by default there, as our ChatGPT review explains.
Where ChatGPT Runs Short for Coding
ChatGPT’s coding limits come as ranges, and heavy Astra use drains Plus quickly, at 5 to 45 messages every five hours. Free and Go get only trial access to Codex. And on the independent index, Codex’s top entry sits five points behind Claude Code’s.
The chat itself is weaker for code than Codex. Plus chat runs GPT-5.6 Sol at up to High effort. The newer GPT-6 Astra and GPT-6.1 Sol reach Plus users only in Codex and ChatGPT Work. Our ChatGPT vs Grok vs DeepSeek comparison also weighs price and data terms.
Who Should Use ChatGPT for Coding, and What to Use Instead
ChatGPT suits developers who want one subscription for chat and a coding agent. Codex’s cost per task on the index is low, the sandbox is on by default, and Plus covers a few sessions a week. It also suits teams already on ChatGPT Business, where code stays out of training by default.
It suits developers who need the top agent score less. Claude Code leads Artificial Analysis’s comparison, as our Claude review covers, though at a much higher cost per task. Grok Build with Grok 4.7 scored 56 on the same index, per our Grok review. Our ChatGPT vs Claude comparison sets the two subscriptions side by side.
How We Research
This guide draws on OpenAI’s own documents, read on September 30, 2026. They include the Codex pricing page, the Codex security docs and the code blocks article.
Agent scores come from Artificial Analysis’s Claude Code vs Codex comparison, read on the same day. We do not run our own coding tests or show sample code, since those would age with every model update. Our criteria live on our about page.
ChatGPT for Coding FAQ
Yes. OpenAI includes Codex in Free, Go, Plus, Pro, Business and Enterprise plans. Free and Go get only limited trial access.
OpenAI estimates 15 to 150 local messages every five hours with GPT-6 Sol, and 5 to 45 with GPT-6 Astra. Weekly limits may apply on top, and Codex shares the allowance with ChatGPT Work.
On Artificial Analysis’s Coding Agent Index, Claude Code with Sonnet 5.5 scores 68 and Codex with GPT-6.1 Sol 63. Codex costs far less per task on that index.
Not by default. OpenAI says the agent runs with network access off, locally and in its cloud containers, unless you enable it.
On personal plans, training is on by default until you opt out in Data controls. Business and Enterprise plans do not train on your code by default.
The Verdict
ChatGPT for coding is worth it on Plus if you want a capable agent inside a general subscription. Codex comes with the plan, starts in a locked-down sandbox and runs cheaply per task. On the independent index, though, Claude Code scores higher. Our page on ChatGPT for productivity covers projects, ChatGPT Work and the always-on dots on Pro.
Keep the sandbox and network defaults, turn off training on personal plans, and watch the weekly cap rather than the five-hour window. If agent scores matter more than cost, compare Claude Code before you pay for Pro. Our page on ChatGPT for marketing covers ChatGPT’s ad platform and the FTC and EU rules on AI content.