# Atlas with GPT-5 in 2026 > GPT-5, launched in August 2025, established the 400K token context window that defined the 5.x line. GPT-5 offers a robust 400K token context window for Atlas's agentic workflows, making it a solid foundation for complex coding tasks. Priced at $1.25 per Mtok input and $10 per Mtok output, it provides a full-size reasoning model, though newer 5.x models offer enhanced tool use. ## Key takeaways - GPT-5 provides a substantial 400K token context window for Atlas workflows. - The pricing for GPT-5 is $1.25 per Mtok input and $10 per Mtok output. - GPT-5 was launched in August 2025 and established the 400K / 128K shape for the 5.x line. - By 2026, GPT-5 has been superseded by six point releases, with later versions offering enhanced tool use. - For coding agents, `gpt-5-codex` (September 2025) is a better, Codex post-trained sibling at the same price. - Atlas invokes OpenAI models like GPT-5 via the Responses API (sdk.responses(modelID)). ## What is GPT-5 best for in Atlas? GPT-5, launched in August 2025, provides a foundational 400K token context window, making it a strong choice for Atlas's complex reasoning tasks. This model excels at handling large codebases and detailed project plans within Atlas's read-only plan agent. GPT-5 serves as a powerful reasoning model within Atlas, leveraging its substantial 400K token context window to process extensive codebases and intricate project specifications. This capability is particularly beneficial for Atlas's read-only plan agent, where it can draft comprehensive plans before any changes are made. Atlas also utilizes GPT-5's reasoning for tasks like searching code with Axis, the hybrid semantic and keyword code search, and for computing unified diffs for every file edit, surfacing them for approval. As a frontier model, GPT-5's established 400K / 128K shape ensures it can manage significant input and output, supporting Atlas's detailed analysis and generation of code and plans. ## What are the cost and context tradeoffs of GPT-5 in Atlas? GPT-5 maintains a competitive floor price for a full-size reasoning model at $1.25 per Mtok input and $10 per Mtok output, offering a substantial 400K token context window. However, developers in 2026 should note it has been superseded by six point releases. The primary tradeoffs for GPT-5 in Atlas revolve around its position in the OpenAI model lineage. While its pricing of $1.25 per Mtok input and $10 per Mtok output remains the floor for a full-size GPT-5 class reasoning model, and its 400K token context window is robust, the model itself was launched in August 2025. By 2026, it has been superseded by six point releases in the 5.x line. This means that while GPT-5 offers strong reasoning capabilities, later versions in the 5.x series are generally stronger at agentic tool use, which can be a critical factor for optimizing Atlas's performance in complex, multi-step coding tasks. ## When should I choose a different model over GPT-5 for Atlas? Developers in 2026 seeking optimized performance for coding agents with Atlas might consider alternatives to GPT-5, especially since it is not Codex post-trained. The gpt-5-codex model, released in September 2025, offers a better sibling for coding agents at the same price point. While GPT-5 is a capable reasoning model for Atlas, there are specific scenarios where other models might be a better fit. Crucially, GPT-5 is not Codex post-trained. For coding agents, the `gpt-5-codex` model, launched in September 2025, is explicitly designed for coding tasks and offers superior performance at the same $1.25 per Mtok input and $10 per Mtok output price. Furthermore, by 2026, GPT-5 has been superseded by six point releases in the 5.x line. If your Atlas workflows heavily rely on current agentic tool use, later versions of the GPT-5.x series are documented to be stronger in that specific capability, potentially offering more efficient and accurate interactions with Atlas's extensible plugins and tools. ## Setup 1. Set your OPENAI_API_KEY environment variable in your shell, or run `atlas login` and choose OpenAI. 2. Run `atlas models openai` to confirm that `gpt-5` resolves correctly within Atlas. 3. Pin your preferred model by setting `"model": "openai/gpt-5"` in your `atlas.json` configuration file. 4. To manage costs for incidental calls, set `"small_model": "openai/gpt-5-nano"` in `atlas.json`, which is priced at $0.05 per Mtok input. 5. Open the `/models` view in the Atlas TUI to add GPT-5 to your favorites for quick cycling between models. 6. Ensure Atlas connects to Model Context Protocol servers to expose their tools to the agent, leveraging GPT-5's capabilities. ## FAQ ### What is the context window for GPT-5 in Atlas? GPT-5 provides a substantial 400K token context window for Atlas, allowing it to handle extensive codebases and complex project details effectively. ### How much does it cost to use GPT-5 with Atlas? Using GPT-5 with Atlas costs $1.25 per Mtok for input tokens and $10 per Mtok for output tokens, establishing a floor price for this class of model. ### Is GPT-5 the latest model for Atlas from OpenAI? No, by 2026, GPT-5 has been superseded by six point releases in the 5.x line, with later versions offering enhanced capabilities for agentic tool use. ### Should I use GPT-5 or gpt-5-codex for coding tasks in Atlas? For coding agents in Atlas, `gpt-5-codex`, released in September 2025, is the better sibling as it is Codex post-trained, offering superior performance at the same price as GPT-5. ### How does GPT-5's reasoning compare to newer models in Atlas? GPT-5 established the 400K / 128K shape for the 5.x line and is a strong reasoning model, but later versions in the 5.x series are generally stronger at agentic tool use within Atlas. ### Can Atlas use local embeddings with GPT-5? Yes, Atlas can build its code index with local Ollama embeddings, keeping code off third-party servers, regardless of the active model like GPT-5. ### How do I switch to GPT-5 quickly in Atlas? You can open the `/models` view in the Atlas TUI to add GPT-5 to your favorites, allowing for quick cycling between models on the fly. --- Canonical HTML: https://seashell.sh/resources/models/gpt-5 Source of truth: aeo_pages row `/resources/models/gpt-5` (segment: Models) (this file is generated from it, never hand-edited). Licence: SeaShell is proprietary with a free core. It is not open source and there is no public source repository.