Models and Reasoning

Choose GPT-6 Astra, GPT-5.6 Sol, Terra, Luna, and reasoning levels by task risk, cost, and latency.

Key takeaways

GPT-6 Astra (gpt-6-astra), released 2026-09-03, is Codex's top model and, from CLI 0.153.4, the bundled default when no model is explicitly configured. It requires CLI 0.153.0 or later.
Astra has a 1,050,000-token context window, 128K max output, API pricing of $10 input / $1 cached input / $50 output per 1M, and an April 30, 2026 knowledge cutoff.
GPT-5.6 Sol remains the strongest GPT-5.6 model; Terra balances everyday engineering and Luna optimizes clear, repeatable, cost-sensitive work.
Reasoning levels are low, medium (the default when unset), high, xhigh, and max. Ultra adds automatic delegation; Max spends more reasoning on one task.
Astra is the first OpenAI model to reach the Critical cybersecurity capability level under the Preparedness Framework. Tighten approvals, sandboxing, and review before granting it high-privilege automation.
GPT-5.4 and GPT-5.4 mini retired from ChatGPT-sign-in Codex on 2026-08-31. GPT-5.5 is a previous-generation regression baseline.
Codex Cloud tasks do not currently let users change the default model, so keep Cloud policy separate.

Current Model Baseline (2026-09-05)

Model	Position	Best fit
GPT-6 Astra (`gpt-6-astra`)	Most capable model, built for the hardest end-to-end work	Long autonomous implementation, large migrations, computer use and browsing, multistep research, document and app output
GPT-5.6 Sol (`gpt-5.6-sol`)	Strongest GPT-5.6 model	Complex coding, computer use, research, cyber, high-risk review
GPT-5.6 Terra (`gpt-5.6-terra`)	Balanced everyday model	Implementation, tests, docs, sustainable team default
GPT-5.6 Luna (`gpt-5.6-luna`)	Fast and affordable	Clear small edits, repeated transformations, bounded subagents
GPT-5.3-Codex-Spark (`gpt-5.3-codex-spark`)	Pro text-only research preview	Very low-latency short iteration

The official models page lists all five as available in the ChatGPT desktop app and web, the Codex CLI, the IDE extension, and Codex cloud, and states that what you actually see depends on the rollout, your sign-in method, and your client. Enterprise workspaces should also check workspace model availability.

GPT-6 Astra Specifications

Item	Value
Model ID	`gpt-6-astra`
Context window	1,050,000 tokens
Max output	128,000 tokens
API pricing (per 1M)	$10 input · $1 cached input · $50 output
Knowledge cutoff	April 30, 2026
Reasoning effort	`low`, `medium`, `high`, `xhigh`, `max`
Supported features	Streaming, structured outputs, function calling, file search, image input, web search, prompt caching
Supported tools	Web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, tool search
Not supported	Realtime, Assistants, Fine-tuning, Embeddings, and audio/video endpoints

How Astra landed in Codex

CLI 0.153.0 introduced Astra support. 0.153.1 added a way to configure Astra through the API without changing the default model or picker visibility. 0.153.3 added it to the Amazon Bedrock Mantle and Runtime pickers. 0.153.4 fixed its visibility in the bundled model picker and made it the bundled default when no model is explicitly configured. Teams that pin model in config.toml are unaffected by that default change, so adopt Astra deliberately.

Critical cybersecurity classification

OpenAI classifies Astra as its first model to reach the Critical cybersecurity capability level under the Preparedness Framework, stating that with the right tools and access it can find previously unknown security flaws and develop new ways to exploit them. Before running Astra with broad automation, narrow /permissions, sandboxing, and approval policy, and re-check auto-approval scope and network access.

August 31 retirement (completed)

GPT-5.4 and GPT-5.4 mini retired from Codex with ChatGPT sign-in on 2026-08-31. Clean up remaining references in config.toml, CI, and team runbooks, moving them to Terra, Luna, or Astra. The retirement does not automatically remove the models from the OpenAI API or API-key-authenticated Codex, so verify those provider catalogs separately.

Operating Rules

Use Astra plus high~max for the hardest and longest work: large-context migrations, multistep research, and tasks mixing computer use with code.
Use Astra or Sol plus high~max and human review for ambiguous or high-blast-radius work.
Use Terra plus medium for everyday engineering; escalate to Sol or Astra when failure cost rises.
Use Luna plus low/medium for explicit, repeatable tasks.
Use Ultra when the goal benefits from automatic delegation, after bounding write scopes and budget.
Use Max for deeper single-task reasoning; do not treat it as an alias for Ultra orchestration.

Astra costs more per token than the GPT-5.6 family. Before raising it to the everyday default, run representative task evals to confirm the quality gain justifies the extra credits, and leave repetitive work on Terra and Luna.

Commands and Configuration

codex --model gpt-6-astra
codex --model gpt-5.6
codex --model gpt-5.6-sol
codex --model gpt-5.6-terra
codex --model gpt-5.6-luna

Authentication, account, and provider rollout can change the exact IDs exposed to a workspace. Use the /model picker as the final availability check.

model = "gpt-6-astra"
model_reasoning_effort = "high"

# Balanced everyday default
# model = "gpt-5.6-terra"
# model_reasoning_effort = "medium"

# Optional dedicated review model
# review_model = "gpt-6-astra"

Provider and Cloud Boundaries

Codex CLI 0.148.0 adds a built-in amazon-bedrock provider configured with AWS profile and region for GPT-5.6 routing. Since 0.153.3, GPT-6 Astra also appears in the Bedrock Mantle and Runtime pickers.
Do not assume direct OpenAI, Bedrock, and custom providers expose identical IDs or reasoning levels; record each provider's /model inventory and eval results.
A local CLI/App/IDE model change does not change the fixed default model of a Codex Cloud task.

Migration Checklist

Upgrade to CLI 0.153.0 or later and confirm with codex --version. Pin model in config.toml if you do not want the 0.153.4 default change to apply.
Review /permissions, sandboxing, and auto-approval scope before adopting Astra, given its Critical cybersecurity classification.
Clean up remaining gpt-5.4 and gpt-5.4-mini references; the ChatGPT-sign-in path retired them on 2026-08-31.
Keep gpt-5.5 only as a regression baseline, and remove gpt-5.2 and gpt-5.3-codex from ChatGPT-sign-in defaults.
Confirm the selected model and reasoning effort with /model and /status.
Compare Astra against Sol and Terra for quality, latency, and credit use on representative team tasks.
Keep Cloud and API-key provider policies separate.

Sources

Models: https://learn.chatgpt.com/docs/models
GPT-6 Astra (API): https://developers.openai.com/api/docs/models/gpt-6-astra
GPT-6 Astra announcement: https://openai.com/index/gpt-6-astra/
Codex CLI: https://developers.openai.com/codex/cli
Slash commands: https://developers.openai.com/codex/cli/slash-commands
Config reference: https://developers.openai.com/codex/config-reference
Changelog: https://learn.chatgpt.com/docs/changelog

Key takeaways

GPT-6 Astra (gpt-6-astra), released 2026-09-03, is Codex's top model and, from CLI 0.153.4, the bundled default when no model is explicitly configured. It requires CLI 0.153.0 or later.
Astra has a 1,050,000-token context window, 128K max output, API pricing of $10 input / $1 cached input / $50 output per 1M, and an April 30, 2026 knowledge cutoff.
GPT-5.6 Sol remains the strongest GPT-5.6 model; Terra balances everyday engineering and Luna optimizes clear, repeatable, cost-sensitive work.
Reasoning levels are low, medium (the default when unset), high, xhigh, and max. Ultra adds automatic delegation; Max spends more reasoning on one task.
Astra is the first OpenAI model to reach the Critical cybersecurity capability level under the Preparedness Framework. Tighten approvals, sandboxing, and review before granting it high-privilege automation.
GPT-5.4 and GPT-5.4 mini retired from ChatGPT-sign-in Codex on 2026-08-31. GPT-5.5 is a previous-generation regression baseline.
Codex Cloud tasks do not currently let users change the default model, so keep Cloud policy separate.

Current Model Baseline (2026-09-05)

Model	Position	Best fit
GPT-6 Astra (`gpt-6-astra`)	Most capable model, built for the hardest end-to-end work	Long autonomous implementation, large migrations, computer use and browsing, multistep research, document and app output
GPT-5.6 Sol (`gpt-5.6-sol`)	Strongest GPT-5.6 model	Complex coding, computer use, research, cyber, high-risk review
GPT-5.6 Terra (`gpt-5.6-terra`)	Balanced everyday model	Implementation, tests, docs, sustainable team default
GPT-5.6 Luna (`gpt-5.6-luna`)	Fast and affordable	Clear small edits, repeated transformations, bounded subagents
GPT-5.3-Codex-Spark (`gpt-5.3-codex-spark`)	Pro text-only research preview	Very low-latency short iteration

GPT-6 Astra Specifications

Item	Value
Model ID	`gpt-6-astra`
Context window	1,050,000 tokens
Max output	128,000 tokens
API pricing (per 1M)	$10 input · $1 cached input · $50 output
Knowledge cutoff	April 30, 2026
Reasoning effort	`low`, `medium`, `high`, `xhigh`, `max`
Supported features	Streaming, structured outputs, function calling, file search, image input, web search, prompt caching
Supported tools	Web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, tool search
Not supported	Realtime, Assistants, Fine-tuning, Embeddings, and audio/video endpoints

How Astra landed in Codex

Critical cybersecurity classification

August 31 retirement (completed)

Operating Rules

Use Astra plus high~max for the hardest and longest work: large-context migrations, multistep research, and tasks mixing computer use with code.
Use Astra or Sol plus high~max and human review for ambiguous or high-blast-radius work.
Use Terra plus medium for everyday engineering; escalate to Sol or Astra when failure cost rises.
Use Luna plus low/medium for explicit, repeatable tasks.
Use Ultra when the goal benefits from automatic delegation, after bounding write scopes and budget.
Use Max for deeper single-task reasoning; do not treat it as an alias for Ultra orchestration.

Commands and Configuration

codex --model gpt-6-astra
codex --model gpt-5.6
codex --model gpt-5.6-sol
codex --model gpt-5.6-terra
codex --model gpt-5.6-luna

Authentication, account, and provider rollout can change the exact IDs exposed to a workspace. Use the /model picker as the final availability check.

model = "gpt-6-astra"
model_reasoning_effort = "high"

# Balanced everyday default
# model = "gpt-5.6-terra"
# model_reasoning_effort = "medium"

# Optional dedicated review model
# review_model = "gpt-6-astra"

Provider and Cloud Boundaries

Codex CLI 0.148.0 adds a built-in amazon-bedrock provider configured with AWS profile and region for GPT-5.6 routing. Since 0.153.3, GPT-6 Astra also appears in the Bedrock Mantle and Runtime pickers.
Do not assume direct OpenAI, Bedrock, and custom providers expose identical IDs or reasoning levels; record each provider's /model inventory and eval results.
A local CLI/App/IDE model change does not change the fixed default model of a Codex Cloud task.

Migration Checklist

Upgrade to CLI 0.153.0 or later and confirm with codex --version. Pin model in config.toml if you do not want the 0.153.4 default change to apply.
Review /permissions, sandboxing, and auto-approval scope before adopting Astra, given its Critical cybersecurity classification.
Clean up remaining gpt-5.4 and gpt-5.4-mini references; the ChatGPT-sign-in path retired them on 2026-08-31.
Keep gpt-5.5 only as a regression baseline, and remove gpt-5.2 and gpt-5.3-codex from ChatGPT-sign-in defaults.
Confirm the selected model and reasoning effort with /model and /status.
Compare Astra against Sol and Terra for quality, latency, and credit use on representative team tasks.
Keep Cloud and API-key provider policies separate.

Sources

Models: https://learn.chatgpt.com/docs/models
GPT-6 Astra (API): https://developers.openai.com/api/docs/models/gpt-6-astra
GPT-6 Astra announcement: https://openai.com/index/gpt-6-astra/
Codex CLI: https://developers.openai.com/codex/cli
Slash commands: https://developers.openai.com/codex/cli/slash-commands
Config reference: https://developers.openai.com/codex/config-reference
Changelog: https://learn.chatgpt.com/docs/changelog

Current Model Baseline (2026-09-05)

GPT-6 Astra Specifications

Operating Rules

Commands and Configuration

Provider and Cloud Boundaries

Migration Checklist

Sources

On this page

Models and Reasoning

Current Model Baseline (2026-09-05)

GPT-6 Astra Specifications

Operating Rules

Commands and Configuration

Provider and Cloud Boundaries

Migration Checklist

Sources

On this page