Atlas with GPT-5.2 Codex is specifically post-trained for long-running software engineering work, excelling at agentic coding tasks within Atlas. It offers a substantial 400K token context window, allowing it to read and rewrite entire services. This specialized capability comes at a cost of $1.75 per million input tokens and $14 per million output tokens, identical to the base GPT-5.2 model.
What is GPT-5.2 Codex best for in Atlas?
GPT-5.2 Codex, released in December 2025, is purpose-built for long-running software engineering tasks within Atlas, excelling at agentic coding. Its post-training specifically targets staying on task across many tool calls, rather than just providing one-shot answers.
Atlas with GPT-5.2 Codex is optimized for complex, multi-step coding workflows. Its specialized training allows Atlas to maintain context and focus across numerous tool invocations, which is critical for tasks like refactoring large services or implementing new features that span multiple files. The model's 400K token context window, combined with Atlas's ability to index code by AST declarations using tree-sitter and search code with hybrid semantic and keyword retrieval, means it can comprehend and modify significant portions of a codebase. Atlas further enhances this by drafting a plan in a read-only plan agent and asking for approval before switching to a build agent, ensuring that GPT-5.2 Codex's agentic capabilities are always permission-gated.
What are the pricing and context window for GPT-5.2 Codex?
GPT-5.2 Codex offers a substantial 400K token context window, enabling Atlas to handle large codebases and complex refactoring tasks. Its pricing is set at $1.75 per million input tokens and $14 per million output tokens, mirroring the base GPT-5.2 model.
The 400K token context window of GPT-5.2 Codex is a significant advantage for developers using Atlas in 2026. This capacity, coupled with a 128K token maximum output, allows Atlas to ingest an entire service's codebase and generate comprehensive rewrites or modifications within a single turn. Crucially, the specialized coding capabilities of GPT-5.2 Codex come at no additional cost compared to the general-purpose GPT-5.2 model, maintaining the same transparent pricing structure of $1.75 per million input tokens and $14 per million output tokens.
When should I choose a different model over GPT-5.2 Codex?
While GPT-5.2 Codex excels at coding, its specialized nature means it is narrower than a general model, making base GPT-5.2 a better default for non-coding reasoning tasks. Furthermore, by February 2026, GPT-5.3 Codex superseded this model at the same price point.
Developers should consider alternatives to GPT-5.2 Codex when their tasks extend beyond pure software engineering. For general reasoning, problem-solving, or tasks requiring broader knowledge, the base GPT-5.2 model is a more appropriate and effective choice. Additionally, it is important to note that GPT-5.2 Codex, released in December 2025, was superseded by GPT-5.3 Codex in February 2026. While both models share the same pricing of $1.75 per million input tokens and $14 per million output tokens, GPT-5.3 Codex offers updated capabilities, making it the preferred option for new projects in 2026 if a specialized coding model is desired.
Setup
- 01Set your OpenAI API key: Ensure the `OPENAI_API_KEY` environment variable is configured, or add your `apiKey` directly under the `openai` provider in your `atlas.json` configuration file.
- 02Confirm model availability: Execute `atlas models openai` in your terminal to verify that `gpt-5.2-codex` is listed and accessible.
- 03Pin the model for the build agent: In your `atlas.json` file, explicitly set `"model": "openai/gpt-5.2-codex"` so Atlas's main build agent utilizes the specialized Codex post-trained weights.
- 04Review all generated actions: Maintain permissions on `ask` for both `bash` and `edit` tools. This ensures every code modification or command generated by GPT-5.2 Codex is presented as a unified diff for your approval before execution.
- 05Enable edit rollback: Configure Atlas to checkpoint edits as git patches. This allows you to easily review and roll back any undesirable outcomes from a GPT-5.2 Codex run.
Frequently asked questions
- What is the context window for GPT-5.2 Codex?
- GPT-5.2 Codex offers a 400K token context window, with a maximum output of 128K tokens, allowing Atlas to process and generate extensive code.
- How much does GPT-5.2 Codex cost in Atlas?
- GPT-5.2 Codex is priced at $1.75 per million input tokens and $14 per million output tokens, identical to the base GPT-5.2 model.
- Is GPT-5.2 Codex good for general reasoning tasks?
- No, GPT-5.2 Codex is narrower than a general model. For non-coding reasoning, the base GPT-5.2 model is a more suitable default choice.
- When was GPT-5.2 Codex released?
- GPT-5.2 Codex was released in December 2025, specifically post-trained for long-running software engineering work.
- Has GPT-5.2 Codex been superseded by a newer model?
- Yes, GPT-5.2 Codex was superseded by GPT-5.3 Codex in February 2026, which offers updated capabilities at the same pricing.
- How does Atlas use GPT-5.2 Codex for code modifications?
- Atlas leverages GPT-5.2 Codex's agentic coding strengths to draft plans in a read-only plan agent, then asks for approval before switching to a build agent. All file edits are surfaced as unified diffs for approval.
- Can Atlas use GPT-5.2 Codex with local embeddings?
- Yes, Atlas can build its code index with local Ollama embeddings, keeping your code off third-party servers, even when using a remote model like GPT-5.2 Codex.
- What are the benefits of GPT-5.2 Codex's post-training for Atlas?
- GPT-5.2 Codex's post-training targets agentic coding, enabling Atlas to stay on task across many tool calls, which is crucial for complex, long-running software engineering projects.
Try SeaShell in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install SeaShellRelated guides
Atlas for Erlang in 2026
Atlas is a terminal-native AI coding agent for Erlang/OTP in 2026. Run it in an app with a rebar.config, map supervisors and gen_server modules, review every diff.
Run the Test Suite and Triage the Failures with Atlas in 2026
How to triage a failing test suite with Atlas in 2026: bash truncates at 2000 lines or 50 KB and saves the full log, then grep groups failures by root cause.
Atlas vs Zed: Terminal AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with Zed, the GPU-accelerated collaborative editor, for developers in 2026. Evaluate their AI models, workflow, and safety features.
Atlas for Electron: Terminal-Native AI Coding for Main, Preload, and Renderer in 2026
Atlas is a terminal-native AI coding agent for Electron in 2026, where the main and renderer split, contextIsolation, and preload bridges are the security model.
Atlas for Rust in 2026
Adopt Atlas, the terminal-native AI coding agent, for Rust development in 2026. Tackle borrow checker errors and clippy lints with Atlas's secure, approval-gated assistance.
Atlas vs PearAI: A Developer's Guide to AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with PearAI, a VS Code fork aggregating open agents, for developers in 2026. Evaluate features, pricing, and ecosystem.
Upgrade a Dependency and Fix the Breakage with Atlas (2026 Workflow)
How to upgrade a dependency and fix the breakage with Atlas in 2026: bash drives the package manager, webfetch pulls the release notes, edit fixes each compiler error.
Automate GitHub Issue and Pull Request Triage with Atlas (2026 Workflow)
How to automate GitHub issue and pull request triage with Atlas in 2026: the atlas github command checks the actor has admin or write permission before it does anything.