GPT-5 Chat is the chat tuned snapshot shipped with the GPT-5 launch, addressed as gpt-5-chat-latest. Unlike later chat variants, GPT-5 Chat kept the full 400K context and 128K max output of the API model, and it costs the same $1.25 per Mtok input and $10 per Mtok output as base GPT-5. Inside Atlas, GPT-5 Chat is the chat flavor without the usual penalty, though it points at a moving snapshot, which makes it a poor choice when you need reproducible runs.
Why does gpt-5-chat-latest keep a 400K window?
GPT-5 Chat is the only chat-latest snapshot in the OpenAI line that retains the full 400K window, since the 5.1 and 5.3 chat variants drop to 128K. GPT-5 Chat shipped alongside the GPT-5 launch and inherited the API model's limits rather than the trimmed ChatGPT ones.
The window difference is the reason to care about GPT-5 Chat at all in 2026. A later chat snapshot at 128K cannot accept the retrieval Atlas produces on a large task, and truncation degrades the answer silently. GPT-5 Chat accepts the whole 400K. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, and that retrieval only pays off if the model can hold what comes back. GPT-5 Chat can.
What is the max output of GPT-5 Chat?
GPT-5 Chat has a 128K max output, versus the 16,384 token cap on the later chat snapshots. That is an eightfold difference, and it means GPT-5 Chat can emit a full rewritten module in one response where a later chat variant would truncate mid file.
A 128K output ceiling matters in Atlas because Atlas computes a unified diff for every file edit and surfaces it for approval before writing, and a substantial refactor produces a substantial diff. GPT-5 Chat, with 128K of output room, does not hit the wall that the 16,384 token chat snapshots hit. Combined with its 400K context, GPT-5 Chat is the only chat tuned OpenAI model that can operate on realistic code at realistic size.
Does GPT-5 Chat cost more than GPT-5?
GPT-5 Chat costs $1.25 per Mtok input and $10 per Mtok output, identical to base GPT-5. There is no price penalty and no price discount for the chat tuning, so the choice between gpt-5-chat-latest and gpt-5 in Atlas is entirely about behavior, not budget.
Identical $1.25 per Mtok input and $10 per Mtok output pricing across GPT-5 Chat and base GPT-5 means the only useful test is empirical. Use /models to compare its behavior against gpt-5 on the same Atlas prompt. Atlas lets you switch the active model and provider on the fly with favorites and recents, so run the same task twice, once on each, and compare the resulting diffs. Atlas snapshots file changes as git patches so edits can be diffed and rolled back, which makes that experiment safe.
Why is a moving snapshot a problem for Atlas?
GPT-5 Chat points at a moving snapshot, which makes it a poor choice when you need reproducible runs. The gpt-5-chat-latest alias tracks whatever OpenAI currently ships under that name, so an Atlas workflow validated last month may behave differently this month with no change on your side.
Reproducibility matters more in an agent than in a chatbot. An Atlas session runs commands, edits files, and creates commits, and a prompt that reliably produced a correct patch is a real asset. Prefer the dated coding models instead when you need a stable snapshot across runs. Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, so a drifting model cannot do damage unreviewed, but it can quietly stop being good at a task you depended on.
Is GPT-5 Chat good enough for the Atlas build agent?
GPT-5 Chat is conversation tuned, so it is less disciplined than gpt-5-codex inside a long tool loop. For the Atlas build agent, which chains dozens of tool calls, gpt-5-codex is the better pin at the same $1.25 per Mtok input and $10 per Mtok output price.
The tradeoff is clean. GPT-5 Chat brings a full 400K window, a 128K output, and conversational fluency. gpt-5-codex brings the same window and output plus post training for exactly the sequential tool calling that Atlas does. If your session is mostly discussion with occasional edits, pin "model": "openai/gpt-5-chat-latest" and enjoy the chat flavor with full context. If your session is mostly building, gpt-5-codex is the disciplined choice for the same money.
Setup
- 01Export OPENAI_API_KEY so the Atlas openai provider can authenticate.
- 02Run `atlas models openai` to confirm gpt-5-chat-latest resolves.
- 03Pin "model": "openai/gpt-5-chat-latest" if you want the chat flavor with full context.
- 04Prefer the dated coding models instead when you need a stable snapshot across runs.
- 05Use /models to compare its behavior against gpt-5 on the same Atlas prompt.
Frequently asked questions
- What is gpt-5-chat-latest?
- GPT-5 Chat is the chat tuned snapshot shipped with the GPT-5 launch. Unlike later chat variants, it kept the full 400K context and 128K max output of the API model, at $1.25 per Mtok input.
- Does gpt-5-chat-latest have a 400K context window?
- Yes. GPT-5 Chat is the only chat-latest snapshot in the line that retains the full 400K window. The 5.1 and 5.3 chat variants drop to 128K.
- How much does GPT-5 Chat cost?
- GPT-5 Chat costs $1.25 per Mtok input and $10 per Mtok output, which is identical to base GPT-5.
- Should I use gpt-5-chat-latest for coding in Atlas?
- For a build agent chaining dozens of tool calls, gpt-5-codex is the better pin at the same price. GPT-5 Chat is conversation tuned, so it is less disciplined inside a long tool loop, though it does keep a full 400K window.
- Why should I avoid a chat-latest alias for reproducible runs?
- GPT-5 Chat points at a moving snapshot, so the model behind gpt-5-chat-latest can change without any change on your side. Prefer the dated coding models when you need a stable snapshot across runs.
- How do I configure Atlas to use gpt-5-chat-latest?
- Export OPENAI_API_KEY, run `atlas models openai` to confirm gpt-5-chat-latest, then pin "model": "openai/gpt-5-chat-latest" in atlas.json if you want the chat flavor with full context.
- How can I compare GPT-5 Chat against GPT-5?
- Use /models to compare its behavior against gpt-5 on the same Atlas prompt. Both cost $1.25 per Mtok input and $10 per Mtok output, so the comparison is purely about behavior.
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 Sourcegraph Cody: Terminal AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with Sourcegraph Cody, an editor extension for whole-codebase search, for developers in 2026. Explore features, pricing, and workflow.
Atlas for Fastify in 2026
Atlas is a terminal-native AI coding agent for Fastify in 2026. It reads plugin encapsulation and JSON schemas, then runs node --test behind a permission prompt.
Run Atlas Headless in CI with Atlas (2026 Workflow)
How to run Atlas headless in CI in 2026: atlas run sends one prompt and exits when the session goes idle, with --format json, --command, and --continue for pipeline steps.
Atlas vs JetBrains Junie in 2026: Terminal Agent vs IDE Debugger-Driving Agent
Atlas vs JetBrains Junie in 2026: Junie drives the IDE debugger and reports 61.6% resolved on SWE-Rebench, bundled from $10/mo. Atlas is a free, open source terminal agent with permission gates.
Atlas for Next.js in 2026
Adopt Atlas, the terminal-native AI coding agent, for Next.js development in 2026. Enhance productivity across App Router, server components, and API routes with secure, reviewable AI assistance.
Atlas for Fortran: fpm.toml, Explicit Interfaces, and fprettify in 2026
Atlas is a terminal-native AI coding agent for Fortran in 2026. It reads modules, explicit interfaces, and intent declarations, runs fpm test behind a prompt, and runs fprettify.
Atlas for Axum in 2026
Atlas is a terminal-native AI coding agent for Axum in 2026. It decodes tower trait-bound errors, adds IntoResponse types, and runs cargo nextest run.