How much does GPT-5 cost per million tokens?

GPT-5 costs $1.25 per Mtok input and $10 per Mtok output. That is still the floor price for a full size GPT-5 class reasoning model.

What is GPT-5's context window?

GPT-5 has a 400K token context window with a 128K max output. That shape, established at the August 2025 launch, held across the 5.x line all the way through GPT-5.3.

How do I use GPT-5 in Atlas?

Set OPENAI_API_KEY in your shell or run `atlas login` and choose OpenAI, run `atlas models openai` to confirm gpt-5 resolves, then pin "model": "openai/gpt-5" in atlas.json.

Is GPT-5 or GPT-5 Codex better for coding?

gpt-5-codex, released September 2025, is the better sibling for coding agents, and it costs the same $1.25 per Mtok input and $10 per Mtok output. GPT-5 is not Codex post trained.

Is GPT-5 still worth using in 2026?

GPT-5 has been superseded by six point releases and later versions are stronger at agentic tool use. GPT-5 remains usable and is priced at the floor for its class, but it is a baseline rather than a recommendation.

What is the cheapest small_model to pair with GPT-5?

Set "small_model": "openai/gpt-5-nano" to keep incidental calls at $0.05 per Mtok input while GPT-5 handles the main agent work at $1.25 per Mtok input.

Does Atlas use the Responses API for GPT-5?

Yes. Atlas invokes OpenAI through sdk.responses(modelID), and GPT-5 is a reasoning model on the Responses API, so its reasoning is preserved across tool results in a long Atlas session.

Atlas with GPT-5: The Original 400K Reasoning Model in 2026

GPT-5 is the August 2025 launch of the GPT-5 line from OpenAI: a 400K token context, a 128K max output, and $1.25 per Mtok input with $10 per Mtok output. Everything from 5.1 through 5.6 is a descendant of this release. Inside Atlas, GPT-5 is a solid, well understood reasoning model on the Responses API, which is exactly how Atlas invokes OpenAI. It is no longer the frontier, but $1.25 in and $10 out remains the floor price for a full size GPT-5 class reasoning model.

What did the August 2025 GPT-5 launch establish?

GPT-5, launched in August 2025, established the 400K context and 128K max output shape that the 5.x line held all the way through GPT-5.3. GPT-5 also set the price floor at $1.25 per Mtok input and $10 per Mtok output, a number later releases matched but did not beat.

Understanding GPT-5 as the reference point explains the whole family. Every later model in the line is described relative to what GPT-5 did: the Codex variants add coding post training at the same price, the mini and nano tiers cut price while keeping the 400K window, and the Pro tiers add reasoning budget at a large premium. When you read that gpt-5-mini is 5x cheaper, the baseline being compared against is GPT-5 at $1.25 per Mtok input and $10 per Mtok output.

How does Atlas invoke GPT-5?

Atlas invokes OpenAI through the Responses API with sdk.responses(modelID), and GPT-5 is a reasoning model on that API. The pairing means GPT-5's reasoning carries forward through the long tool chains an Atlas session produces, rather than being discarded after each tool result.

Atlas's normal mode is a chain: read a file, search, propose an edit, run a command, read the output, adjust. GPT-5 as a reasoning model on the Responses API keeps its chain of thought coherent across that sequence. Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, so a human approval sits between many of those steps, and preserved reasoning is what stops GPT-5 from re-deriving its analysis from scratch after every approval.

Is a 400K context enough for a whole repository?

GPT-5 provides a 400K token context and a 128K output ceiling, enough to hold a large service and its tests rather than an entire monorepo. Atlas indexes code by AST declarations using tree-sitter, not blind line windows, so the 400K on GPT-5 fills with whole symbols rather than truncated slices.

A 400K window on GPT-5 is a retrieval budget, not a repository dump. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, which is what makes the budget go far: the files that reach GPT-5 are the ones that bear on the change, including call sites you would have forgotten. Because GPT-5 bills $1.25 per Mtok input, what fills that window is also what you pay for, so precise retrieval is simultaneously a quality and a cost lever.

Should you use GPT-5 or GPT-5 Codex in Atlas?

Use gpt-5-codex over GPT-5 for coding agents. GPT-5 is not Codex post trained, and gpt-5-codex, released September 2025, is the better sibling for coding agents at the same $1.25 per Mtok input and $10 per Mtok output price. The specialization costs nothing.

The choice is free on cost, so it turns entirely on the task. GPT-5 remains the better general reasoner for design discussion, analysis, and writing. gpt-5-codex is post trained for the long sequential tool calling that is Atlas's normal mode. If you are pinning one model as the Atlas build agent and your work is code, gpt-5-codex is the better pin. Atlas lets you switch the active model and provider on the fly with favorites and recents, so keeping both in /models favorites is the practical setup.

Is GPT-5 obsolete in 2026?

GPT-5 has been superseded by six point releases, and later versions are stronger at agentic tool use. In 2026, GPT-5 is a baseline rather than a recommendation, though its $1.25 per Mtok input and $10 per Mtok output remains the floor price for a full size GPT-5 class reasoning model.

There is a narrow but real case for GPT-5 in 2026: reproducibility and familiarity. GPT-5 is a fixed, well characterized model that many teams have already tuned prompts against, and its 400K context and 128K output are unchanged. If your Atlas workflows were built and validated on GPT-5, moving is a decision, not a default. Atlas snapshots file changes as git patches so edits can be diffed and rolled back, which makes A/B testing GPT-5 against a newer release on real tasks safe to do.

Setup

01Set OPENAI_API_KEY in your shell, or run `atlas login` and choose OpenAI.
02Run `atlas models openai` and confirm gpt-5 resolves.
03Pin "model": "openai/gpt-5" in atlas.json.
04Set "small_model": "openai/gpt-5-nano" to keep incidental calls at $0.05 per Mtok input.
05Open /models in the TUI to add GPT-5 to favorites for quick cycling.

Frequently asked questions

How much does GPT-5 cost per million tokens?: GPT-5 costs $1.25 per Mtok input and $10 per Mtok output. That is still the floor price for a full size GPT-5 class reasoning model.
What is GPT-5's context window?: GPT-5 has a 400K token context window with a 128K max output. That shape, established at the August 2025 launch, held across the 5.x line all the way through GPT-5.3.
How do I use GPT-5 in Atlas?: Set OPENAI_API_KEY in your shell or run `atlas login` and choose OpenAI, run `atlas models openai` to confirm gpt-5 resolves, then pin "model": "openai/gpt-5" in atlas.json.
Is GPT-5 or GPT-5 Codex better for coding?: gpt-5-codex, released September 2025, is the better sibling for coding agents, and it costs the same $1.25 per Mtok input and $10 per Mtok output. GPT-5 is not Codex post trained.
Is GPT-5 still worth using in 2026?: GPT-5 has been superseded by six point releases and later versions are stronger at agentic tool use. GPT-5 remains usable and is priced at the floor for its class, but it is a baseline rather than a recommendation.
What is the cheapest small_model to pair with GPT-5?: Set "small_model": "openai/gpt-5-nano" to keep incidental calls at $0.05 per Mtok input while GPT-5 handles the main agent work at $1.25 per Mtok input.
Does Atlas use the Responses API for GPT-5?: Yes. Atlas invokes OpenAI through sdk.responses(modelID), and GPT-5 is a reasoning model on the Responses API, so its reasoning is preserved across tool results in a long Atlas session.

Try Atlas in your terminal

The terminal-native AI coding agent. Free core, single binary.

Install Atlas

Related 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.

Atlas vs Qodo: Choosing Your AI Coding Agent in 2026

Comparing Atlas, the terminal-native AI coding agent, with Qodo 2.0, the multi-agent PR reviewer, for developers in 2026. Evaluate features, pricing, and workflow.

Atlas for Fiber in 2026

Atlas is a terminal-native AI coding agent for Fiber in 2026. It knows fasthttp reuses buffers, tests handlers with app.Test(), and diffs every edit first.

Migrate a Deprecated API Across Every Callsite with Atlas (2026 Workflow)

How to migrate a deprecated API across every callsite with Atlas in 2026: the lsp tool's findReferences enumerates callers, todowrite tracks them, apply_patch migrates each one.

Atlas vs v0: A Developer's Guide for 2026

Comparing Atlas and v0 in 2026: Atlas offers terminal-native AI coding with explicit diffs and BYO model keys, while v0 provides a visual editor for React/Next.js apps.

Atlas for Scala in 2026

Atlas is a terminal-native AI coding agent for Scala in 2026. Run it in a project with a build.sbt, let it read your traits and implicits, and approve every diff.

Atlas vs Claude Code: Terminal AI Coding Agents in 2026

Compare Atlas and Claude Code in 2026 for terminal AI coding. Atlas offers free core, BYO model keys, and robust diffing, while Claude Code provides deep Claude tuning and strong multi-step execution.

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.

Browse this resource hub