GPT-5 Codex arrived in September 2025 as the first Codex post trained model in the GPT-5 family. It carries the same 400K context, 128K output, and $1.25 per Mtok input with $10 per Mtok output pricing as GPT-5, tuned specifically for long agentic coding sessions. Inside Atlas, GPT-5 Codex behaves better than base GPT-5 when a task needs dozens of sequential tool calls, which is Atlas's normal mode. The coding specialization costs nothing extra.
What made GPT-5 Codex different from GPT-5 in September 2025?
GPT-5 Codex, released September 2025, was the first Codex post trained model in the GPT-5 family. GPT-5 Codex keeps the 400K context, the 128K output, and the $1.25 per Mtok input pricing of GPT-5, and adds coding specific post training at zero price premium over the base model.
The September 2025 release created the pattern that every later Codex model followed: same limits, same price, different training objective. GPT-5 Codex was tuned specifically for long agentic coding sessions rather than one shot answers. For a team evaluating models for Atlas, that means the decision between GPT-5 Codex and base GPT-5 has never been a budget decision. Both bill $1.25 per Mtok input and $10 per Mtok output. It is purely a question of whether the session is code work.
Why does GPT-5 Codex handle dozens of sequential tool calls better?
GPT-5 Codex behaves better than base GPT-5 when a task needs dozens of sequential tool calls, which is Atlas's normal mode. A single Atlas task routinely reads files, searches, runs a build, reads the failure, and patches, and GPT-5 Codex was post trained against exactly that long agentic shape.
Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, multiplying the tool calls further. Base GPT-5 tends to lose the thread across a long chain: it restates the problem, forgets an earlier constraint, or drifts to a different fix. GPT-5 Codex holds the goal. Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, so a persistent model is safe to run persistently: you still see and approve each step.
Can GPT-5 Codex rewrite a whole service in one turn?
GPT-5 Codex has a 400K context with a 128K output ceiling, sized for read-a-service, rewrite-a-service turns. Atlas indexes code by AST declarations using tree-sitter, not blind line windows, so the 400K window GPT-5 Codex receives is composed of whole declarations rather than arbitrary line ranges.
The read-a-service, rewrite-a-service turn is the shape GPT-5 Codex was built for. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, so the retrieval that fills GPT-5 Codex's window pulls in the interfaces, the tests, and the call sites together. GPT-5 Codex can then emit the complete rewritten module inside its 128K output ceiling, and Atlas computes a unified diff for every file edit and surfaces it for approval before writing.
How do you keep a long GPT-5 Codex session reviewable?
Keep a long GPT-5 Codex session reviewable in Atlas by leaving permissions on ask for edit and bash, so each Codex write lands as a reviewable unified diff. Atlas reads git branches, status, and diffs, and can stage and create commits on your behalf, so a long GPT-5 Codex session produces clean, attributable history.
GPT-5 Codex is post trained to keep working across dozens of tool calls, which means an unreviewed GPT-5 Codex session can produce a large, tangled changeset. Atlas git-aware commits are the answer: let GPT-5 Codex work, and let Atlas stage and commit at meaningful boundaries so the history explains itself afterward. Atlas also snapshots file changes as git patches so edits can be diffed and rolled back, which turns a bad GPT-5 Codex run into a revert rather than a manual cleanup.
Is GPT-5 Codex outdated compared to newer Codex models?
GPT-5 Codex is three Codex generations old: gpt-5.1-codex, gpt-5.2-codex, and gpt-5.3-codex all followed the September 2025 release. In 2026, GPT-5 Codex is the oldest member of a well developed line, and newer Codex models are the stronger picks on capability.
GPT-5 Codex is also coding specialized, so it is not the right pick for general reasoning or writing tasks: base GPT-5 is the better generalist at the same $1.25 per Mtok input and $10 per Mtok output. The narrow, honest case for GPT-5 Codex in 2026 is a team already standardized on it with tuned prompts and validated workflows. Atlas lets you switch the active model and provider on the fly with favorites and recents, so comparing GPT-5 Codex against a newer Codex generation on a real task takes one keystroke.
Setup
- 01Export OPENAI_API_KEY so the Atlas openai provider can authenticate.
- 02Run `atlas models openai` and confirm gpt-5-codex resolves.
- 03Pin "model": "openai/gpt-5-codex" in atlas.json.
- 04Keep Atlas permissions on ask for edit and bash so each Codex write lands as a reviewable unified diff.
- 05Use Atlas git-aware commits so a long Codex session produces clean, attributable history.
Frequently asked questions
- What is GPT-5 Codex?
- GPT-5 Codex is the first Codex post trained model in the GPT-5 family, released September 2025. It keeps the 400K context, 128K output, and $1.25 per Mtok input pricing of GPT-5, tuned for long agentic coding sessions.
- Does GPT-5 Codex cost more than GPT-5?
- No. GPT-5 Codex costs $1.25 per Mtok input and $10 per Mtok output, exactly the same as base GPT-5. The coding specific post training carries zero price premium.
- How do I set gpt-5-codex as the Atlas model?
- Export OPENAI_API_KEY, run `atlas models openai` to confirm gpt-5-codex, then pin "model": "openai/gpt-5-codex" in atlas.json so the build agent uses the Codex weights.
- Is GPT-5 Codex good for non coding tasks?
- No. GPT-5 Codex is coding specialized, so it is not the right pick for general reasoning or writing tasks. Base GPT-5 is the better generalist and costs the same $1.25 per Mtok input.
- Is GPT-5 Codex still competitive in 2026?
- GPT-5 Codex is three Codex generations old, with gpt-5.1-codex, gpt-5.2-codex, and gpt-5.3-codex all following it. It remains capable at $1.25 per Mtok input, but newer Codex models are stronger.
- How do I review what GPT-5 Codex changed in my repo?
- Keep Atlas permissions on ask for edit and bash so each Codex write lands as a reviewable unified diff, and use Atlas git-aware commits so a long GPT-5 Codex session produces clean, attributable history.
- What context window does gpt-5-codex have?
- GPT-5 Codex has a 400K token context with a 128K output ceiling, the same limits as base GPT-5, sized for read-a-service, rewrite-a-service turns.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for SolidJS: A Terminal-Native AI Coding Agent Built on Solid, for Solid, in 2026
Atlas is a terminal-native AI coding agent for SolidJS in 2026. It reads your createSignal and createMemo graph, fixes destructured props that break reactivity, and runs vitest.
Atlas vs Gemini CLI: A Developer's Guide to Terminal AI Agents in 2026
Atlas vs Gemini CLI in 2026: Compare terminal AI coding agents. Atlas offers permission-gated tool calls and diff-based approvals. Gemini CLI provides a million-plus token context and a free tier.
Atlas for Haskell in 2026
Atlas is a terminal-native AI coding agent for Haskell in 2026. Run it in a project with a .cabal file or a stack.yaml, let cabal build drive it, and review the diff.
Atlas vs Greptile: Terminal AI Coding Agents in 2026
Comparing Atlas and Greptile in 2026. Atlas offers terminal-native AI coding with permission-gated tools. Greptile reviews code with sandbox execution, catching 20% more bugs.
Atlas for TypeScript in 2026
In 2026, TypeScript developers leverage Atlas, the terminal-native AI coding agent, to enhance productivity. Atlas understands your types, ensures code quality, and offers robust safety features.
Atlas for Nuxt: Auto-Imports, useAsyncData, and Nitro Handlers in 2026
Atlas is a terminal-native AI coding agent for Nuxt in 2026. It reads nuxt.config.ts, pages/ routes, composables/ auto-imports, and server/api/ Nitro handlers, and tests with @nuxt/test-utils.
Atlas vs Windsurf: Terminal AI Coding Agents in 2026
Atlas and Windsurf offer distinct AI coding experiences in 2026. Atlas provides a terminal-native TUI with permission-gated tools, while Windsurf features an in-editor live preview and a dedicated GUI.
Atlas for Assembly: Registers, Calling Conventions, and nasm in 2026
Atlas is a terminal-native AI coding agent for Assembly in 2026. It reads .asm and .S sources, tracks System V and AAPCS64 calling conventions, and assembles with nasm behind a prompt.