GPT-5.6 is OpenAI's July 2026 flagship with a 1.05M token window (1,050,000 tokens), priced at $5 / $30 per Mtok (input / output) with 128K max output tokens. Atlas routes OpenAI through the Responses API rather than chat completions, so GPT-5.6 reasoning state persists across tool calls in a session. The cost warning is blunt: $30 per Mtok output is six times Claude Sonnet 5's $10 at a similar context window, and reasoning tokens are billed as output.
Why does Atlas route GPT-5.6 through the Responses API?
Atlas calls sdk.responses(modelID) for the OpenAI provider, which preserves reasoning items between tool calls instead of discarding them each turn. For GPT-5.6, driven in July 2026 as OpenAI's flagship, that means reasoning state persists across a session rather than restarting at every tool result.
The difference shows up in agentic loops, not in one-shot prompts. A chat completions call throws away the model's reasoning trace once the turn ends, so a model that spent effort deducing something has to deduce it again after the next tool result comes back. Because Atlas uses the Responses API path via sdk.responses(modelID), GPT-5.6 keeps that trace. In a session where Atlas permission-gates every tool call and fans out subagents, persisting reasoning is the difference between an agent that builds a mental model and one that keeps starting over.
How big is the GPT-5.6 context window?
GPT-5.6 has a 1.05M token context window, exactly 1,050,000 tokens. That is the largest in the GPT-5 family and larger than Gemini 3.1 Pro's 1,048,576. GPT-5.6 also carries 128K max output tokens, so a large multi-file patch fits in a single Atlas edit turn.
A 1,050,000 token window changes what you can hand the model. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion and indexes by AST declarations using tree-sitter, so a broad retrieval returns many whole declarations rather than fragments. GPT-5.6 can hold all of them plus the running transcript. The 128K max output tokens is the other half: Atlas computes a unified diff for every file edit and surfaces it for approval, and a model that can emit the whole patch avoids stitching turns together.
How much does GPT-5.6 cost per million tokens?
GPT-5.6 costs $5 / $30 per Mtok (input / output). The output side is the one to watch: $30 per Mtok output is six times Claude Sonnet 5's $10 at a similar context window, and reasoning tokens are billed as output, so real cost on hard tasks runs well above the headline rate.
Headline pricing understates GPT-5.6 in an agent. Reasoning tokens bill as output, and a long Atlas session on a hard problem generates a lot of reasoning between tool results, especially with the Responses API preserving and extending that trace. Budget accordingly. If the work is exploratory or mechanical, the $30 per Mtok output rate is money spent on thinking you did not need. If the work is a genuinely hard debugging or design problem, that same reasoning is the entire product you are buying.
When should you pick a different model than GPT-5.6 in Atlas?
Switch away from GPT-5.6 whenever output volume dominates. Claude Sonnet 5 charges $10 per Mtok output against GPT-5.6's $30 at a similar context window, so a session that generates long diffs or long summaries costs three times as much on GPT-5.6 for no clear gain.
Model choice in Atlas is not a commitment. Atlas lets you switch the active model and provider on the fly with favorites and recents, so the practical move is to keep GPT-5.6 starred for hard reasoning and drive something cheaper the rest of the time. Watch for the specific failure mode: a task where the model reasons at length and then emits a small edit is a task where you are paying $30 per Mtok for tokens you never read. GPT-5.6 is worth it when the reasoning is the answer.
How do you set up Atlas to run on GPT-5.6?
Export OPENAI_API_KEY=sk-... or run atlas login and choose OpenAI, then confirm the model list with atlas models openai. In the TUI, run /models, pick OpenAI, then GPT-5.6. To make it the session default, pin "model": "openai/gpt-5.6" in atlas.json.
Atlas resolves OpenAI models from a registry, so atlas models openai is the check that tells you whether GPT-5.6 is actually reachable with your key before you build a session around it. Credentials work either way: the environment variable is fine for a scratch shell, and atlas login stores the key in the credential store instead. The atlas.json pin is optional. If you only want GPT-5.6 for a stretch of work, the /models dialog changes the active model without touching config.
Setup
- 01Export `OPENAI_API_KEY=sk-...`, or run `atlas login` and choose OpenAI.
- 02Confirm the model list: `atlas models openai`.
- 03Run `/models` in the TUI, pick OpenAI, then GPT-5.6.
- 04Optional: pin it with `"model": "openai/gpt-5.6"` in atlas.json.
- 05Budget for reasoning tokens, which are billed as output at $30 per Mtok.
Frequently asked questions
- what is the context window of gpt-5.6
- GPT-5.6 has a 1.05M token context window, exactly 1,050,000 tokens. That is the largest in the GPT-5 family and larger than Gemini 3.1 Pro's 1,048,576. Max output is 128K tokens.
- how much does gpt-5.6 cost per million tokens
- GPT-5.6 costs $5 / $30 per Mtok (input / output). Reasoning tokens are billed as output, so real cost on hard tasks runs well above the headline rate.
- does atlas use the responses api for openai models
- Yes. Atlas calls `sdk.responses(modelID)` for the OpenAI provider rather than chat completions, so GPT-5.6 reasoning state persists across tool calls in a session.
- how do i use gpt-5.6 in atlas
- Export `OPENAI_API_KEY=sk-...` or run `atlas login` and choose OpenAI. Confirm with `atlas models openai`, then run `/models` in the TUI and pick OpenAI, then GPT-5.6.
- gpt-5.6 vs claude sonnet 5 cost
- GPT-5.6 charges $30 per Mtok output, six times Claude Sonnet 5's $10 at a similar context window. On input, GPT-5.6 is $5 per Mtok against Sonnet 5's $2.
- how do i make gpt-5.6 the default model in atlas
- Pin `"model": "openai/gpt-5.6"` in atlas.json. For a single session, use the `/models` dialog in the TUI instead.
- why is my gpt-5.6 bill higher than the listed price
- GPT-5.6 bills reasoning tokens as output at $30 per Mtok, so a hard task that reasons at length costs well above the headline rate even if the visible answer is short.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Sourcery: Terminal AI Coding Agents in 2026
Compare Atlas and Sourcery for 2026. Atlas offers terminal-native AI coding with permission-gated tools and local embeddings. Sourcery provides Python-first LLM review and rules-based refactoring, starting at
Atlas for R: A Terminal-Native AI Coding Agent for tidyverse, roxygen2, and testthat in 2026
Atlas is a terminal-native AI coding agent for R in 2026. It reads roxygen2 docblocks and renv.lock, rewrites loops as dplyr or purrr pipelines, and runs devtools::test().
Atlas vs Fine.dev: Terminal AI Coding Agents in 2026
Compare Atlas and Fine.dev for terminal AI coding in 2026. Atlas offers a free core TUI with permission-gated tools, while Fine.dev provides asynchronous cloud agents and prebuilt libraries.
Atlas for COBOL: Copybooks, PIC Clauses, and GnuCOBOL in 2026
Atlas is a terminal-native AI coding agent for COBOL in 2026. It reads your divisions and copybooks, explains a paragraph before touching it, and compiles with cobc under GnuCOBOL.
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 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.
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 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.