GPT-5.1 Codex mini is the fast tier of OpenAI's November 2025 Codex family, billing $0.25 per Mtok input, $2 per Mtok output while retaining a 400K tokens window. Within Atlas, GPT-5.1 Codex mini earns its slot through breadth rather than depth: it powers parallel subagents, bulk sweeps across a repository, session titles, and transcript compaction, all at a rate low enough that you stop rationing them.
Why run parallel Atlas subagents on GPT-5.1 Codex mini?
Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and GPT-5.1 Codex mini makes that breadth affordable at $0.25 per Mtok input. Six explorers sweeping a repository at once becomes routine rather than a budgeting conversation.
Subagents perform real tool work, not conversation, which is why Codex post training matters even in the fast tier. GPT-5.1 Codex mini is meaningfully better at tool driven edits than a general purpose model at the same rate, so parallel explorers come back with usable findings instead of confident guesses. Background sessions let the sweep proceed while you keep working in the foreground, and the bill stays small enough that a wasted branch of exploration costs nothing worth mourning.
Does GPT-5.1 Codex mini keep a full 400K window?
GPT-5.1 Codex mini retains a 400K tokens window plus a 128K output ceiling, which is unusual for a fast tier. Most inexpensive models amputate context first, dropping to 32K or less. GPT-5.1 Codex mini instead trims reasoning depth and leaves the window untouched.
That retained window is what makes bulk sweeps possible at $0.25 per Mtok input, $2 per Mtok output. Atlas indexes code by AST declarations using tree-sitter, not blind line windows, so a broad pass across a repository arrives as complete declarations rather than mangled halves of a function. Hand GPT-5.1 Codex mini an enormous slice of a codebase and ask one mechanical question spanning all of it. A rival capped at 32K simply cannot attempt the job, however clever it may be.
What mechanical jobs suit GPT-5.1 Codex mini?
GPT-5.1 Codex mini handles well specified mechanical jobs, so pinning it as the main Atlas model is reasonable for renames, test scaffolding, and config conversions. At $0.25 per Mtok input, an entire day of tedious editing produces a bill you barely notice.
Draw the line at judgment. Renaming a symbol across forty files, scaffolding tests against an interface that already exists, or migrating a config format all sit comfortably within reach, and Atlas computes a unified diff for every file edit and surfaces it for approval before writing, so nothing lands unseen. Designing a caching layer or chasing a race condition through three services does not, because GPT-5.1 Codex mini is weaker on genuinely hard architectural reasoning than codex or codex-max.
Can a cheap model like GPT-5.1 Codex mini cost more?
Yes. GPT-5.1 Codex mini is cheap enough to over use, and a wrong mini run that burns tool calls can cost more than one correct larger run. A $0.25 per Mtok input rate never guarantees the session ends up cheaper overall.
Spend follows turn count, not sticker rate. Handed a job beyond its depth, GPT-5.1 Codex mini thrashes: it opens irrelevant files, ships a patch that breaks the build, reads the breakage, ships another patch. Every wasted turn re-bills the whole accumulated transcript. Watch the cost footer, and once GPT-5.1 Codex mini has failed the same check twice, move up rather than funding a third attempt at the fast tier rate.
Is GPT-5.1 Codex mini the right Atlas small_model?
GPT-5.1 Codex mini fits the small_model slot well, where it writes session titles, compacts long transcripts, and backs subagent fan out. Configure it once and the incidental traffic Atlas generates all day runs at $0.25 per Mtok input rather than full rate.
Every session quietly produces work nobody requested. Conversations need naming, long transcripts need summarising before they overflow, and subagents need weights to run on. Paying a frontier rate for that housekeeping is waste. GPT-5.1 Codex mini is cheap without being incompetent, and because the summarizer only needs to read a lot and write a little, the fast tier is a natural fit. Optionally pin it as the main model too when the day is mechanical.
How do you sweep a whole repository with GPT-5.1 Codex mini?
Sweeping a repository with GPT-5.1 Codex mini means pointing its 400K tokens window at a broad slice and asking one narrow question repeatedly. Inventory every deprecated helper, list untested exports, flag stale imports: breadth first questions that reward a wide window and a $0.25 per Mtok input rate.
Sweeps of this shape reward volume over insight, which is exactly the shape the fast tier was built for. Split the repository into slices, dispatch one background explorer per slice, and let the results accumulate while you continue working in the foreground. Because a sweep reads enormously and writes almost nothing, the $2 per Mtok output rate barely registers, and the whole inventory lands for less than a single careful turn on a frontier tier would have cost you.
Setup
- 01Export OPENAI_API_KEY.
- 02Confirm with `atlas models openai`.
- 03Set "small_model": "openai/gpt-5.1-codex-mini" so Atlas subagents, titles, and summaries run cheap.
- 04Optionally pin it as the main "model" for mechanical work like renames and test scaffolding.
- 05Fan out parallel Atlas subagents against it, since $0.25 per Mtok input makes wide exploration affordable.
Frequently asked questions
- What does gpt-5.1-codex-mini cost per million tokens?
- GPT-5.1 Codex mini from OpenAI bills $0.25 per Mtok input, $2 per Mtok output, placing it in the fast tier of the November 2025 Codex family.
- Is the gpt-5.1-codex-mini context window smaller?
- No. GPT-5.1 Codex mini retains a 400K tokens window with a 128K output ceiling, which is unusual for an inexpensive fast tier model.
- Which model should fill the Atlas small_model slot?
- Set "small_model": "openai/gpt-5.1-codex-mini" so Atlas subagents, titles, and summaries run cheap. Its Codex post training beats a general model at the same $0.25 per Mtok input on tool driven edits.
- Is gpt-5.1-codex-mini good for running many agents at once?
- Yes. Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and $0.25 per Mtok input makes wide exploration across many of them affordable.
- Can gpt-5.1-codex-mini handle refactoring?
- Mechanical refactoring such as renames and test scaffolding suits it, and you may pin it as the main model. Genuinely hard architectural reasoning does not, since GPT-5.1 Codex mini is weaker there than codex or codex-max.
- Does a cheaper model always mean a cheaper session?
- No. GPT-5.1 Codex mini is cheap enough to over use, and a wrong mini run that burns tool calls can cost more than one correct larger run. Spend tracks turn count, not the sticker rate.
- How do I check gpt-5.1-codex-mini is available in Atlas?
- Export OPENAI_API_KEY, then confirm with `atlas models openai`. The command prints the resolved OpenAI models, and gpt-5.1-codex-mini should appear in the list.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for React in 2026
Adopt Atlas, the terminal-native AI coding agent, for React development in 2026. Enhance your workflow with intelligent code search, refactoring, and testing for React components and hooks.
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.
Debug a Single Failing Test with Atlas in 2026
How to debug one failing test with Atlas in 2026: run it in isolation with bash, walk the call graph with the lsp tool, and fix the code, not the assertion.
Atlas for Godot: Terminal-Native AI Coding for GDScript and the Node Tree in 2026
Atlas is a terminal-native AI coding agent for Godot in 2026, working across GDScript, the node tree, and signals, where scene structure is half the program.
Atlas for PowerShell: A Terminal-Native AI Coding Agent for Modules and Pester in 2026
Atlas is a terminal-native AI coding agent for PowerShell in 2026. It reads your .psm1 module and .psd1 manifest, adds SupportsShouldProcess, and runs Invoke-Pester behind a prompt.
Atlas for Gin in 2026
Atlas is a terminal-native AI coding agent for Gin in 2026. It reads router groups and binding tags, then runs go test ./... -race behind a permission prompt.
Atlas for Java in 2026
Adopt Atlas, the terminal-native AI coding agent, for Java development in 2026. Enhance your workflow with intelligent code search, refactoring, and robust safety features for Maven and Gradle projects.
Atlas vs Traycer in 2026: Terminal Agent That Writes Code vs a Planning Layer Above One
Atlas vs Traycer in 2026. Traycer plans and verifies but writes no code itself, so you pay it on top of an agent. Atlas plans and writes, free and open source.