GPT-5.2 is the December 2025 GPT-5 refresh from OpenAI. It kept the 400K tokens context and 128K max output of GPT-5.1 but repriced upward to $1.75 per Mtok input, $14 per Mtok output. Inside Atlas, the terminal-native AI coding agent, GPT-5.2 is the general purpose sibling of GPT-5.2 Codex, and its 400K window with a 128K output ceiling is a workable ratio for read-many, write-large agent turns. The honest caveat is that GPT-5.2 is not Codex post trained, so for pure agentic coding gpt-5.2-codex is usually the better pick at identical price.
What is the 400K context and 128K output ratio of GPT-5.2 good for?
GPT-5.2 pairs a 400K tokens context with a 128K max output ceiling, which is a workable ratio for read-many, write-large agent turns. The December 2025 refresh kept both numbers from GPT-5.1, so the shape of an agent turn on GPT-5.2 is unchanged.
A 400K read budget against a 128K write budget is roughly three to one, and that ratio suits how Atlas actually works. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, pulling a broad set of relevant declarations into the GPT-5.2 window, then Atlas computes a unified diff for every file edit and surfaces it for approval before writing. The 128K output ceiling of GPT-5.2 means that diff does not need splitting for most tasks. Atlas indexes code by AST declarations using tree-sitter, not blind line windows, so the 400K tokens fill with whole functions and types rather than truncated fragments.
How much did GPT-5.2 cost compared to GPT-5.1?
GPT-5.2 is 40 percent more expensive per input token than GPT-5.1, $1.75 versus $1.25 per Mtok, for the same 400K window. The December 2025 refresh repriced upward without changing the context window or the 128K max output ceiling.
The price move on GPT-5.2 is worth naming plainly because nothing about the specs justifies it on paper. Same 400K tokens context, same 128K max output, 40 percent more per input token, and $14 per Mtok on output. For an Atlas user the practical effect is that read heavy sessions cost more than they did on GPT-5.1 for identical work. The mitigation is the same discipline that helps on any model: let Atlas hybrid semantic and keyword search choose the files, and let the Atlas plan agent, which drafts a plan in a read-only mode and asks before switching to a build agent, decide what actually needs reading before GPT-5.2 sees a single token.
Should you use GPT-5.2 or GPT-5.2 Codex for coding in Atlas?
Use GPT-5.2 Codex for pure agentic coding. GPT-5.2 is not Codex post trained, and gpt-5.2-codex is usually the better pick at identical price, sharing the same key and the same 400K window, so there is no cost or context penalty for switching.
GPT-5.2 is the general purpose sibling of GPT-5.2 Codex, and the two are designed to be swapped. Same API key, same 400K tokens window, same price of $1.75 per Mtok input and $14 per Mtok output. What differs is post training: GPT-5.2 Codex was trained toward software engineering work, and GPT-5.2 was not. That makes GPT-5.2 the right choice when a session mixes coding with general reasoning, documentation, or analysis, and makes gpt-5.2-codex the right choice when the session is purely agentic coding. Atlas lets you switch the active model and provider on the fly with favorites and recents, so use /models to compare GPT-5.2 head to head with gpt-5.2-codex on the same repo task.
How do you keep cheap calls off GPT-5.2 output pricing in Atlas?
Set "small_model": "openai/gpt-5-nano" in atlas.json so cheap calls do not bill at the $14 per Mtok output rate of GPT-5.2. Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and each of those turns bills separately.
Incidental traffic is where a GPT-5.2 bill quietly grows. Session titles, summaries, and small mechanical lookups do not need a 400K token reasoning model, and every one of them that lands on GPT-5.2 pays $14 per Mtok output. Pointing the small_model slot at openai/gpt-5-nano routes that traffic away permanently while leaving "model": "openai/gpt-5.2" in charge of the actual build loop. GPT-5.2 is a reasoning model on the Responses API, which is how Atlas calls every OpenAI model through sdk.responses, so its reasoning trace persists across the tool loop and it does not waste output re deriving context it already had.
Setup
- 01Set OPENAI_API_KEY in the environment.
- 02Run `atlas models openai` to confirm gpt-5.2 resolves from models.dev.
- 03Pin "model": "openai/gpt-5.2" in atlas.json.
- 04Set "small_model": "openai/gpt-5-nano" so cheap calls do not bill at $14 per Mtok output.
- 05Use /models to compare GPT-5.2 head to head with gpt-5.2-codex on the same repo task.
Frequently asked questions
- what is the context window of gpt-5.2
- GPT-5.2 has a 400K tokens context window with a 128K max output ceiling. The December 2025 refresh kept both numbers from GPT-5.1 and changed only the price.
- how much does gpt-5.2 cost per million tokens
- GPT-5.2 costs $1.75 per Mtok input and $14 per Mtok output. Input is 40 percent more expensive than GPT-5.1, which billed at $1.25 per Mtok for the same 400K window.
- gpt-5.2 vs gpt-5.2-codex for coding
- GPT-5.2 is not Codex post trained, so for pure agentic coding gpt-5.2-codex is usually the better pick at identical price. Both share the same key and the same 400K window, so switching costs nothing.
- how do I set gpt-5.2 as the model in atlas
- Set OPENAI_API_KEY in the environment, run `atlas models openai` to confirm gpt-5.2 resolves from models.dev, then pin "model": "openai/gpt-5.2" in atlas.json.
- what small_model should I use with gpt-5.2 in atlas
- Set "small_model": "openai/gpt-5-nano" so cheap calls do not bill at the $14 per Mtok output rate of GPT-5.2. Titles, summaries, and small lookups do not need the main reasoning tier.
- does gpt-5.2 keep reasoning across tool calls in atlas
- Yes. GPT-5.2 is a reasoning model on the Responses API, which is how Atlas calls every OpenAI model through sdk.responses. The reasoning trace persists across the tool loop.
- what is the max output of gpt-5.2
- GPT-5.2 supports a 128K max output ceiling, carried over from GPT-5.1. Against its 400K tokens context, that gives roughly a three to one read to write ratio.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for dbt: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for dbt. Read the ref() DAG, convert a table model to incremental, run dbt build against dev, and add tests in 2026.
Atlas vs Kiro in 2026: Terminal Agent Compared to AWS's Spec-Driven IDE and CLI
Atlas vs Kiro in 2026. Kiro writes EARS-notation specs before code and charges credits; Atlas is a free, open source terminal agent with diff-before-write review.
Atlas for Laravel in 2026
Atlas is a terminal-native AI coding agent for Laravel in 2026. Run it in a Laravel app with an artisan file, add a controller or Pest test, and review the diff.
Plan a Multi-File Change Before Editing with Atlas in 2026
How to plan a multi-file change with Atlas in 2026: the plan agent denies all edit tools, you research with codebase_search and lsp, then plan_exit hands off.
Add a Regression Test for a Bug Fix with Atlas in 2026
How to add a regression test with Atlas in 2026: red first, then green. bash records the exit code, write creates the failing test, and edit applies the fix.
Atlas for Dart in 2026
Adopt Atlas, the terminal-native AI coding agent, for Dart development in 2026. Enhance productivity with intelligent code search, refactoring, and robust safety features across your Dart projects.
Atlas vs Fine.dev: Terminal AI Coding Agents in 2026
Compare Atlas and Fine.dev for terminal AI coding in 2026. Atlas offers a free core TUI with permission-gated tools, while Fine.dev provides asynchronous cloud agents and prebuilt libraries.
Atlas for Expo: Terminal-Native AI Coding for expo-router and Config Plugins in 2026
Atlas is a terminal-native AI coding agent for Expo apps in 2026, covering expo-router file routes, config plugins, and EAS build profiles with diff-first review.