Codex Models

Q: What is the default model in Codex?

GPT-6.1 Sol (gpt-6.1-sol) is the default model in Codex as of Nov 6, 2026, set in ~/.codex/config.toml and shared by the desktop app, CLI and IDE extension.

Q: How do I switch models in Codex CLI?

Start Codex with codex -m gpt-6.1-terra (or another model name), or type /model during a session. GPT-6.1 Terra needs Codex CLI 0.159.0 or later.

As of Nov 6, 2026 Codex recommends four models: GPT-6 Astra (most capable), GPT-6.1 Sol (the default, near Astra at lower cost), GPT-6.1 Terra (fast, for writing and computer use) and GPT-6 Luna (most efficient). They replaced a line of coding-tuned models that ran from codex-1 in 2025 through GPT-5-Codex, GPT-5.1-Codex-Max and GPT-5.3-Codex.

0:04 The best model for Codex as of Nov 6, 2026 is GPT-6 Astra for the hardest work and GPT-6.1 Sol for nearly everything else: Sol is the default, gives "near-Astra performance for complex work at a lower cost", and gets about three times as many messages on Plus (Models, pricing). GPT-6.1 Terra is the fast choice for clear writing, images and computer use, and GPT-6 Luna the cheapest for simple, high-volume tasks and the only model on ChatGPT Free and Go. Switch with /model in the CLI, codex -m <model>, or the model control under the composer in the app and IDE.

1:10 Codex no longer has models of its own. From 2025 to early 2026 OpenAI shipped coding-tuned versions of each model (codex-1, GPT-5-Codex, GPT-5.1-Codex-Max, GPT-5.2-Codex, GPT-5.3-Codex); since GPT-5.4 in March 2026 Codex has used the same general models as ChatGPT Work. The lineage is below. Every model OpenAI makes is on ChatGPT Models.

Quick facts

Default model	GPT-6.1 Sol (`gpt-6.1-sol`), as of Nov 6, 2026
Most capable	GPT-6 Astra (`gpt-6-astra`)
Fastest for writing and computer use	GPT-6.1 Terra (`gpt-6.1-terra`)
Most efficient	GPT-6 Luna (`gpt-6-luna`)
Sol context window in Codex	500,000 tokens by default
Free and Go	GPT-6 Luna only, in the desktop app
First Codex model	codex-1, May 16, 2025
Last "-Codex" model	GPT-5.3-Codex, Feb 5, 2026

Which model should I use in Codex?

Model	OpenAI's description	Plus messages per five hours	Credits per 1M tokens (in / out)	API price per 1M tokens (in / out)
GPT-6 Astra	"Our most capable model for complex work across code, apps, and research"	5 to 45	250 / 1,250	$10 / $50
GPT-6.1 Sol	"Near-Astra performance for complex work at a lower cost than Astra"	15 to 160	50 / 250	$2 / $10
GPT-6.1 Terra	"A fast, efficient model for clear writing, visual understanding, and computer use"	35 to 285	25 / 125	$1 / $5
GPT-6 Luna	"Our most efficient model for focused, high-volume tasks"	350 to 3,000	2.5 / 12.5	$0.10 / $0.50

Sources: descriptions from Models; message ranges (local messages, Plus and Standard Business) and credits from Pricing; API prices (Standard, short context) from OpenAI API pricing; all read Nov 6, 2026.

A simple rule from those numbers: start with Sol; move to Astra when Sol gets stuck on a hard, multi-step change; drop to Luna for renames, small fixes and bulk edits where messages matter more than depth; use Terra when the task is mostly writing, reading images, or driving apps with Computer Use. OpenAI publishes no head-to-head coding scores among these four in its Codex docs; benchmark results are collected on OpenAI Benchmarks, and comparisons with rival agents on Codex vs Claude Code.

Which plans get which models?

Free and Go: GPT-6 Luna at Standard speed in the desktop app, subject to rollout (pricing).
Plus: GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna, and "limited Astra usage" with credits for more (ChatGPT Work and Codex).
Pro $100 and $200, Business Premium seats: the full allowance can go to Astra. Pro $500: also Astra Ultrafast and GPT-6.1 Sol Ultrafast (ChatGPT Work and Codex, About ChatGPT Pro tiers).
Enterprise and Edu: GPT-6.1 Sol is off until an administrator turns it on (pricing).

GPT-6.1 Sol and Terra run in Codex and ChatGPT Work but not in ordinary ChatGPT chat, and Terra does not support UAE inference residency (pricing). The IDE extension does not list GPT-6.1 Terra as of Nov 6, 2026 (Models).

Reasoning effort, Max and Ultra

0:51 Each model can think for longer or shorter. Effort runs Light (called Low in the CLI), Medium, High and Extra High; Max spends more time on a single task, and Ultra splits the work across subagents running in parallel. GPT-6 Luna goes up to Max but not Ultra. In the desktop app, a Power control offers presets such as Luna High, GPT-6 Sol Light and Medium, and Astra Light, Medium and Extra High (some paid plans omit Astra Extra High); Advanced picks a model and effort directly (Models). Higher effort usually gives better answers on complex work but takes longer and uses more of your limit.

How do I change the model?

CLI: codex --model gpt-6.1-sol (or -m), codex exec -m gpt-6.1-sol "Review the current changes", or /model in a session. GPT-6.1 Terra needs CLI 0.159.0 or later; GPT-6 Astra needs 0.153.0 or newer.
Default for every surface: model = "gpt-6.1-sol" in ~/.codex/config.toml, which the desktop app, CLI and IDE extension share (Models).
App and IDE: the model and reasoning control below the composer.

With an API key, Codex can use whichever API models your key reaches (pricing). Custom providers must offer a Responses API endpoint, since current Codex releases do not support Chat Completions-only endpoints (Models).

Codex model history

Date	Model	What it was
Aug 10, 2021	OpenAI Codex	GPT-3 descendant that turned language into code; powered GitHub Copilot; deprecated March 2023
May 16, 2025	codex-1	A version of OpenAI o3 tuned for software engineering; ran the first cloud Codex
May 16, 2025	codex-mini-latest	A version of o4-mini for the Codex CLI, $1.50 / $6 per 1M tokens in the API
Sep 15, 2025	GPT-5-Codex	GPT-5 tuned for agentic coding; worked "independently for more than 7 hours"; API on Sep 23, 2025
Nov 7, 2025	GPT-5-Codex-Mini	About 4 times more usage than GPT-5-Codex
Nov 13, 2025	GPT-5.1-Codex, GPT-5.1-Codex-Mini	The GPT-5.1 generation
Nov 18, 2025	GPT-5.1-Codex-Max	Introduced the Extra High (`xhigh`) effort
Dec 18, 2025	GPT-5.2-Codex	API on Jan 14, 2026
Feb 5, 2026	GPT-5.3-Codex	"Runs 25% faster for Codex users"; the first model OpenAI treated as a high security capability under its Preparedness Framework
Feb 12, 2026	GPT-5.3-Codex-Spark	Pro research preview with Cerebras, over 1,000 tokens per second, text only, 128k context
Mar 5, 2026	GPT-5.4	First general model with native computer use; experimental 1M context
Mar 17, 2026	GPT-5.4 mini	Used "30% as much" of the limits as GPT-5.4
Apr 23, 2026	GPT-5.5	Arrived with browser use and automatic approval reviews
Jul 9, 2026	GPT-5.6 Sol, Terra, Luna	Launched with ChatGPT Work; Sol at medium effort was the default
Week of Aug 31, 2026	GPT-6 Astra	Reasoning, computer use and "stronger judgment"
Sep 22, 2026	GPT-6 Sol, GPT-6 Luna	Cheaper per token than the GPT-5.6 models
Sep 29, 2026	GPT-6.1 Sol	The default as of Nov 6, 2026
Oct 26, 2026	GPT-6.1 Terra	Fast model for writing, images and computer use

Sources: OpenAI Codex (Aug 10, 2021), Introducing Codex (May 16, 2025), Introducing upgrades to Codex (Sep 15, 2025), Changelog and What's new, read Nov 6, 2026.

Retired Codex models

Retirements apply only when you sign in with ChatGPT; API keys keep the models the API still offers (Models).

The older gpt-5.2-codex, gpt-5.1-codex family, gpt-5.1 and gpt-5 left the picker on Apr 7, 2026 and were gone on Apr 14, 2026 (changelog).
GPT-5.3-Codex and GPT-5.2 were deprecated for ChatGPT sign-in on May 26, 2026 (changelog).
GPT-5.4 and GPT-5.4 mini were retired on Aug 31, 2026 (replace with gpt-6-sol and gpt-6-luna); GPT-5.3-Codex-Spark on Sep 14, 2026 (Models).
GPT-5.5 was retired on Oct 14, 2026 at noon PDT (19:00 UTC); OpenAI points Plus, Pro, Business, Enterprise and Edu to GPT-6.1 Sol and Free and Go to GPT-6 Luna (Models).

In the API, gpt-5-codex shut down on Jul 23, 2026, and gpt-5.3-codex (still $1.75 / $14 per 1M tokens) is due to shut down on Apr 1, 2027, with gpt-6-sol as its replacement (OpenAI deprecations, API pricing). The full list of retired ChatGPT models is on Retired ChatGPT Models.

Special-purpose models

Codex's rate card also lists models most people never pick: GPT-5.6 Sol at promotional pricing "at least through November 21, 2026", Daybreak Blue and Daybreak Red for approved security defenders (they need "Trusted Access for Cyber"), and GPT-Rosalind-Research for life sciences (pricing). See GPT-Rosalind and Cyber Models and Codex Security.

Frequently asked questions

What is the best model for Codex? As of Nov 6, 2026 GPT-6 Astra is OpenAI's most capable model in Codex, and GPT-6.1 Sol, the default, comes close at a lower cost with about three times the messages on Plus. Most users should start with GPT-6.1 Sol.

What is the default model in Codex? GPT-6.1 Sol (gpt-6.1-sol) is the default model in Codex as of Nov 6, 2026, set in ~/.codex/config.toml and shared by the desktop app, CLI and IDE extension.

Is GPT-5-Codex still available? Not with a ChatGPT sign-in. GPT-5-Codex and its successors up to GPT-5.3-Codex have been retired from Codex for ChatGPT accounts, and gpt-5-codex shut down in the API on Jul 23, 2026.

Which Codex model do Free users get? ChatGPT Free and Go get GPT-6 Luna in the Codex view of the desktop app, at Standard speed and subject to rollout.

How do I switch models in Codex CLI? Start Codex with codex -m gpt-6.1-terra (or another model name), or type /model during a session. GPT-6.1 Terra needs Codex CLI 0.159.0 or later.