GPT-5.5 is the April 2026 OpenAI release that pushed the GPT-5 line to a 1.05M tokens (1,050,000) context, up from the 400K carried by GPT-5 through GPT-5.3. Pricing is $5 per Mtok input, $30 per Mtok output. Inside Atlas, the terminal-native AI coding agent, GPT-5.5 is the model you pick when a task genuinely needs 2.6x the window that GPT-5.2 and GPT-5.3 offered. The catch is that the larger window came with a price rise: input went from $1.75 per Mtok on GPT-5.3 to $5 per Mtok on GPT-5.5, so the big context is emphatically not free.
What changed in GPT-5.5 compared with GPT-5.3?
GPT-5.5 was the first non Pro GPT-5 with a 1,050,000 token window, a 2.6x jump over the 400K carried by GPT-5.2 and GPT-5.3. The April 2026 release retained the 128K max output of the 5.x line, so the read side grew while the write side stayed put.
The headline change in GPT-5.5 is context, not output. Where GPT-5.2 and GPT-5.3 topped out at 400K tokens, GPT-5.5 reads up to 1,050,000, which is what makes a genuine multi service sweep possible in a single Atlas turn. Max output stayed at 128K, retained from the 5.x line, so the size of a patch GPT-5.5 can emit is unchanged. In Atlas that pairing works well because Atlas computes a unified diff for every file edit and surfaces it for approval before writing: you want to read broadly and write in reviewable chunks. Atlas indexes code by AST declarations using tree-sitter, not blind line windows, so the declarations entering the GPT-5.5 window are whole ones.
How much does the bigger GPT-5.5 context window actually cost?
GPT-5.5 raised input price to $5 per Mtok, up from $1.75 on GPT-5.2 and GPT-5.3, so the 1,050,000 token window is not free. A 1M token input load on GPT-5.5 bills $5 before a single output token, and output runs at $30 per Mtok.
The price story on GPT-5.5 is the honest counterweight to the context story. Nearly tripling the input rate means a habit of dumping the repo into the window is now roughly three times as expensive per token as it was on GPT-5.3, on top of being 2.6x larger. The right Atlas discipline with GPT-5.5 is to use Atlas hybrid code search rather than dumping the repo, since 1M input tokens bills at $5. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, so the files that land in the GPT-5.5 window are the ones the task needs. Use the big window as a ceiling for hard cases, not as a default fill level.
How do you fix GPT-5.5 not appearing in Atlas?
Run `atlas providers` if GPT-5.5 does not appear in the model list, which usually means the models.dev cache is stale. Confirm the model with `atlas models openai` after OPENAI_API_KEY is set in your shell, then pin "model": "openai/gpt-5.5" in atlas.json.
Model resolution problems in Atlas are almost always a resolution issue, not a key issue. Atlas resolves model ids from models.dev, and a cached list from before the April 2026 GPT-5.5 release will not contain gpt-5.5. Checking `atlas providers` is the documented step when the model does not appear. Once GPT-5.5 shows up in `atlas models openai`, pinning it in atlas.json makes it the main loop model. Atlas lets you switch the active model and provider on the fly with favorites and recents, so you can also select GPT-5.5 through the TUI without editing config at all, which is the faster path when you only want it for one task.
When should you pick GPT-5.6 instead of GPT-5.5?
Pick GPT-5.6 in most cases, because GPT-5.6 shipped in July 2026 and supersedes GPT-5.5 at the same $5 per Mtok input and $30 per Mtok output price. Newer at identical cost is a straightforward call unless you need a pinned GPT-5.5 snapshot.
There is no cost argument for staying on GPT-5.5 in 2026, which is the plain truth of it. GPT-5.6 carries the same price point, so the reasons to keep GPT-5.5 pinned come down to reproducibility, a benchmarking baseline, or a config you do not want to touch mid project. On the other side, if your work fits comfortably inside 400K tokens, the older GPT-5.2 and GPT-5.3 bill input at $1.75 per Mtok against the $5 of GPT-5.5, which is a real saving for tasks that never needed the larger window. GPT-5.5 is a reasoning model driven through the Responses API in Atlas, so tool heavy sessions keep their reasoning trace regardless of which of these you pick.
Setup
- 01Set OPENAI_API_KEY in your shell.
- 02Confirm the model resolves with `atlas models openai`.
- 03Pin "model": "openai/gpt-5.5" in atlas.json.
- 04Use Atlas hybrid code search rather than dumping the repo, since 1M input tokens bills at $5.
- 05Check `atlas providers` if the model does not appear, which usually means the models.dev cache is stale.
Frequently asked questions
- what is the context window of gpt-5.5
- GPT-5.5 has a 1.05M token context window (1,050,000 tokens). Released in April 2026, it was the first non Pro GPT-5 with that window, a 2.6x jump over the 400K carried by GPT-5.2 and GPT-5.3.
- how much does gpt-5.5 cost per million tokens
- GPT-5.5 costs $5 per Mtok input and $30 per Mtok output. Input rose from $1.75 per Mtok on GPT-5.2 and GPT-5.3, so the larger window came with a real price increase.
- gpt-5.5 not showing up in atlas models list
- Check `atlas providers`. If GPT-5.5 does not appear after running `atlas models openai`, it usually means the models.dev cache is stale rather than an OPENAI_API_KEY problem.
- how do I set gpt-5.5 as the model in atlas.json
- Set OPENAI_API_KEY in your shell, confirm with `atlas models openai`, then pin "model": "openai/gpt-5.5" in atlas.json. You can also select it mid session from the Atlas TUI.
- is gpt-5.5 or gpt-5.6 better for atlas in 2026
- GPT-5.6 shipped in July 2026 and supersedes GPT-5.5 at the same $5 / $30 price, so GPT-5.6 is the default choice. Keep GPT-5.5 pinned only for reproducibility or a fixed benchmarking baseline.
- does the 1m token window on gpt-5.5 cost extra
- Filling the 1,050,000 token window of GPT-5.5 bills $5 per Mtok input, so a 1M token load costs $5 before any output. Use Atlas hybrid code search instead of dumping the repo into the window.
- what is the max output of gpt-5.5
- GPT-5.5 retains the 128K max output of the 5.x line. The April 2026 release grew the context window to 1,050,000 tokens but left the output ceiling unchanged.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for Rust in 2026
Adopt Atlas, the terminal-native AI coding agent, for Rust development in 2026. Tackle borrow checker errors and clippy lints with Atlas's secure, approval-gated assistance.
Atlas vs Gemini CLI: A Developer's Guide to Terminal AI Agents in 2026
Atlas vs Gemini CLI in 2026: Compare terminal AI coding agents. Atlas offers permission-gated tool calls and diff-based approvals. Gemini CLI provides a million-plus token context and a free tier.
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 for FastAPI in 2026
Atlas is a terminal-native AI coding agent for FastAPI in 2026. It reads Pydantic v2 models and Depends(), then runs pytest behind a permission prompt.
Atlas for Haskell in 2026
Atlas is a terminal-native AI coding agent for Haskell in 2026. Run it in a project with a .cabal file or a stack.yaml, let cabal build drive it, and review the diff.
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.
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 Polars: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for Polars. Build LazyFrame chains, push scan_csv predicates into the reader, and read explain() plans in 2026.