Codex Models: The Full 2026 List

As of July 24, 2026, the current generation in Codex is the GPT-5.6 family, three tiers that went GA on July 9, 2026. They share the same generation number; the name is the capability tier within it. GPT-5.6 Sol is the flagship and the only tier with max reasoning effort and ultra mode (native parallel subagents). GPT-5.6 Terra is the everyday workhorse, roughly GPT-5.5 performance at about half the price. GPT-5.6 Luna is the cheap, fast tier for mechanical work.

You do not have to choose at all: Codex auto-selects by default, and the default tracks the recommended model, so you are already on the current generation. Pick manually only when you want to control the cost-capability trade-off per session.

The complete list of OpenAI models currently selectable in Codex, newest first, with prices per million tokens verified July 24, 2026:

The row that decides most setups: GPT-5.6 Terra delivers roughly GPT-5.5 performance at about half GPT-5.5's price ($2.5/$15 vs $5/$30 per MTok), which is why it is the rational default for routine work. The full benchmark breakdown and a step-by-step decision guide are in best Codex model.

Beyond the current generation, a few older and specialty models still appear in the picker, per OpenAI's Codex docs:

You can switch models per session from the Codex model picker, or set a persistent default in the config file. Pair the model with a reasoning effort level, the two knobs together decide cost and depth:

# ~/.codex/config.toml
model = "gpt-5.6-terra"
model_reasoning_effort = "high"

Reasoning effort ranges over minimal, low, medium, high, and xhigh, with a max level that only GPT-5.6 Sol supports. Turn it up for planning and hard debugging; turn it down for mechanical edits. The full explanation is in best Codex model.

Short version: default to GPT-5.6 Terra ($2.5 in / $15 out per MTok) for routine work, escalate to GPT-5.6 Sol ($5 in / $30 out per MTok) for hard or long agentic runs where ultra mode and max effort earn their 2x price, and drop to GPT-5.6 Luna ($1 in / $6 out per MTok) for cheap mechanical tasks. If you would rather not think about it, Codex's auto-select is a fine choice.

The tier system only pays off if switching is cheap. AIDEN orchestrates your Codex CLI alongside Claude Code on one kanban board, with per-story model choice, so Sol can take the hard migration while Terra handles routine features and a Claude model takes whichever story suits it. Every model across both vendors is on the models hub, and the head-to-head with running Codex bare is in AIDEN vs Codex CLI.