GPT-5.5 Pro is OpenAI's April 2026 maximum-effort reasoning tier at $30 / $180 per Mtok (input / output), on the full 1.05M token window (1,050,000 tokens) with 128K output. This is a model you invoke deliberately for one hard question, not one you leave selected in an Atlas session. At $180 per Mtok output it is 18 times Claude Sonnet 5's $10, and latency is measured in minutes on hard prompts.
What is GPT-5.5 Pro actually for inside Atlas?
GPT-5.5 Pro is a model you invoke deliberately for one hard question, not one you leave selected in an Atlas session. OpenAI's April 2026 maximum-effort reasoning tier runs the highest reasoning effort OpenAI exposes, on the full 1,050,000 token window, and it bills accordingly.
Think of GPT-5.5 Pro as a consultation, not a driver. The question that justifies it looks like: why does this cache invalidation race only fire under load, or what is the right decomposition for this service before we touch it. You bring GPT-5.5 Pro the full context, take the answer, and leave. Atlas lets you switch the active model and provider on the fly with favorites and recents, which is what makes a single-question invocation practical rather than a session-long commitment.
How does GPT-5.5 Pro pair with Atlas plan mode?
GPT-5.5 Pro is well suited to Atlas plan mode: run one expensive read-only planning pass, then hand the plan to a cheap build agent. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, so the $30 / $180 per Mtok turn cannot touch a file.
The read-only guarantee is what makes an expensive planning model rational. The plan agent cannot write, so the most costly turn in the session is also the safest one. GPT-5.5 Pro has 128K output tokens, enough for a long architectural writeup with the diff sketched inline, and the 1,050,000 token window means the whole affected surface can be resident while it reasons. Then the build agent, running on something ordinary, executes a plan that has already been thought through.
How much does GPT-5.5 Pro cost per million tokens?
GPT-5.5 Pro costs $30 / $180 per Mtok (input / output). The output rate is the one that decides your behavior: $180 per Mtok output is 18 times Claude Sonnet 5's $10, and reasoning tokens bill as output, so a maximum-effort turn is expensive by construction.
Reasoning tokens billing as output is not a footnote at this tier, it is the bill. GPT-5.5 Pro is the highest reasoning effort OpenAI exposes, which means it generates a lot of reasoning, which means it generates a lot of billed output even when the visible answer is a few paragraphs. That is the product you are buying and it is priced honestly. Just do not buy it by accident by leaving GPT-5.5 Pro selected while Atlas runs routine retrieval turns.
Why does GPT-5.5 Pro latency break an interactive Atlas session?
GPT-5.5 Pro latency is measured in minutes on hard prompts, which makes an interactive TUI loop feel broken. Atlas is a terminal-native TUI rendered with SolidJS through the OpenTUI renderer, and a terminal loop that stalls for minutes per turn stops being a conversation.
Latency is the practical reason GPT-5.5 Pro cannot be a session default, separate from the money. The Atlas rhythm is fast: retrieve, propose, approve at the permission gate, apply, check. Minutes-long turns destroy that rhythm even for trivial requests, because the model applies maximum effort whether or not the question warrants it. Use GPT-5.5 Pro where a multi-minute wait is proportionate to the stakes, which is exactly the one hard question you invoked it for.
How do you switch to GPT-5.5 Pro and back in Atlas?
Set OPENAI_API_KEY or run atlas login, verify with atlas models openai, then switch to GPT-5.5 Pro with /models only for a specific planning or debugging question. Switch back with model.cycle_recent as soon as the answer lands, so the rest of the session does not bill at the Pro rate.
The switch-back discipline is the whole operating procedure for GPT-5.5 Pro. Atlas keeps favorites and recents, and model.cycle_recent flips you between the last two models without reopening the dialog, so returning from the $30 / $180 per Mtok tier to your everyday driver is one keystroke. Build the habit of cycling back immediately, before the next routine turn. A GPT-5.5 Pro session that you forgot to leave is the most expensive way to run Atlas.
Setup
- 01Set `OPENAI_API_KEY` or run `atlas login`.
- 02Verify: `atlas models openai`.
- 03Switch to it with `/models` only for a specific planning or debugging question.
- 04Switch back with `model.cycle_recent` as soon as the answer lands, so the rest of the session does not bill at the Pro rate.
- 05Use Atlas plan mode for the expensive pass, then hand the plan to a cheap build agent.
Frequently asked questions
- how much does gpt-5.5 pro cost per million tokens
- GPT-5.5 Pro costs $30 / $180 per Mtok (input / output). The $180 output rate is 18 times Claude Sonnet 5's $10, and reasoning tokens bill as output.
- what is gpt-5.5 pro's context window
- GPT-5.5 Pro runs on the full 1.05M token window (1,050,000 tokens) with 128K output tokens.
- should i leave gpt-5.5 pro selected in atlas
- No. GPT-5.5 Pro is a model you invoke deliberately for one hard question, not one you leave selected. Switch back with `model.cycle_recent` as soon as the answer lands.
- why is gpt-5.5 pro so slow
- GPT-5.5 Pro applies the highest reasoning effort OpenAI exposes, so latency is measured in minutes on hard prompts. That makes an interactive TUI loop feel broken if it is your default model.
- how do i use gpt-5.5 pro for planning in atlas
- Run one expensive read-only planning pass on GPT-5.5 Pro, then hand the plan to a cheap build agent. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent.
- how do i switch to gpt-5.5 pro in atlas
- Set `OPENAI_API_KEY` or run `atlas login`, verify with `atlas models openai`, then switch with `/models` for the specific question you need it for.
- gpt-5.5 pro vs claude sonnet 5 cost
- GPT-5.5 Pro charges $180 per Mtok output, 18 times Claude Sonnet 5's $10. Sonnet 5 is the everyday driver, and GPT-5.5 Pro is the deliberate one-question escalation.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
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.
Atlas for Nim: A Terminal-Native AI Coding Agent for Nimble Packages and Macros in 2026
Atlas is a terminal-native AI coding agent for Nim in 2026. It reads .nimble requires and asterisk-exported symbols, adds std/unittest suites, runs nimble test, formats with nph.
Atlas for SQL in 2026
Atlas is a terminal-native AI coding agent for SQL in 2026. Run it in a repo with your migrations or .sql files, optimize a query, and review the diff before applying.
Atlas for Phoenix in 2026
Atlas is a terminal-native AI coding agent for Phoenix in 2026. It reads contexts, LiveView modules, and Ecto changesets, then runs mix test behind a prompt.
Onboard to an Unfamiliar Codebase with Atlas in 2026
How to onboard to an unfamiliar codebase with Atlas in 2026: use codebase_search, glob, read, lsp, task, and todowrite to build a mental model fast.
Diagnose a Hanging or Long-Running Command with Atlas in 2026
How to diagnose a hanging command with Atlas in 2026: the bash tool races every command against a timeout and tells you whether it is slow or blocked on input.
Atlas for TensorFlow: Terminal-Native AI Coding for Keras 3 and tf.data in 2026
Atlas is a terminal-native AI coding agent for TensorFlow and Keras 3 in 2026, where tf.function graph tracing and tf.data pipelines are what make training fast.
Atlas vs Codebuff: Terminal AI Coding Agents in 2026
Atlas and Codebuff are terminal AI coding agents for 2026. Compare Atlas's terminal-native TUI, permission-gated tools, and diff review with Codebuff's multi-agent system and flexible pricing.