# Atlas with GPT-5 Nano in 2026 > GPT-5 Nano offers a remarkable 400K token context window at just $0.05 per Mtok input, making it the cheapest model in the OpenAI registry. In 2026, Atlas developers can integrate OpenAI's GPT-5 Nano to power highly cost-effective operations. This model is ideal for Atlas's small_model slot, handling session titles, summaries, and cheap subagent fan out, thanks to its remarkable $0.05 per Mtok input pricing and a substantial 400K token context window. It offers significant savings while maintaining crucial context awareness. ## Key takeaways - GPT-5 Nano is the cheapest OpenAI model in 2026 at $0.05 per Mtok input. - It features a substantial 400K token context window, excellent for bulk scanning. - Ideal for Atlas's `small_model` slot: session titles, summaries, and subagent fan out. - Not suitable as the main agent for multi step refactors; quality degrades before 128K max output. - 25x cheaper than GPT-5 and 5x cheaper than gpt-5-mini for input tokens. - Configure it in `atlas.json` for the `small_model` slot, not the main `model` slot. ## What is GPT-5 Nano best for in Atlas? GPT-5 Nano excels as Atlas's small_model, providing an incredibly cost-effective solution for specific tasks in 2026. With input pricing at just $0.05 per Mtok, it is 25x cheaper than GPT-5, making it perfect for generating session titles, summaries, and fanning out work to subagents. This model is specifically designed to fill the `small_model` slot within Atlas, where its primary role is to manage less computationally intensive but context-aware operations. Atlas leverages GPT-5 Nano for tasks such as creating concise session summaries, generating descriptive session titles, and efficiently distributing work to parallel background subagents. Its 400K token context window ensures that even these "small" tasks benefit from a broad understanding of the codebase or ongoing session, without incurring the higher costs associated with larger, more powerful models. This strategic placement allows developers to optimize their operational expenses significantly. ## What are the cost and context window benefits of GPT-5 Nano? GPT-5 Nano stands out in 2026 for its exceptional value, offering a 400K token context window at an input price of only $0.05 per Mtok. This makes it the cheapest model in the entire OpenAI registry, providing a 25x cost reduction compared to GPT-5 and 5x compared to gpt-5-mini. The pricing structure of GPT-5 Nano is its most compelling feature for Atlas users. At $0.05 per Mtok for input and $0.40 per Mtok for output, it represents a significant cost saving, particularly for high-volume operations like bulk file scanning or extensive summarization. Despite its low cost, the model retains a substantial 400K token context window, which is remarkable and allows Atlas to maintain a deep understanding of the project state or large code files. This combination of affordability and a generous context window makes GPT-5 Nano an ideal choice for tasks where broad context is needed but the highest reasoning capabilities are not critical. ## When should I choose a different model over GPT-5 Nano for Atlas? While GPT-5 Nano offers incredible value, developers in 2026 should avoid using it as the main agent for complex, multi step refactors. Its quality degrades long before reaching its 128K max output, and it will lose the thread on intricate tasks, necessitating a stronger tier model for core development. GPT-5 Nano is not intended to be the primary driving force for Atlas's main agent, especially when tackling sophisticated, multi step refactoring projects or intricate code generation. The model's reasoning capabilities, while present, are not sufficient for maintaining coherence and accuracy across extended, complex operations. Although it boasts a 128K max output on paper, practical experience shows that output quality diminishes significantly well before this limit is reached. For critical tasks that demand robust reasoning, precise code modifications, or sustained conversational depth, Atlas users should configure a stronger, higher-tier model in the main `model` slot within their `atlas.json` configuration. This ensures that the build agent and other core functions receive the necessary intelligence to perform reliably. ## Setup 1. Set your OPENAI_API_KEY environment variable. 2. Run `atlas models openai` and confirm `gpt-5-nano` is available. 3. Add `"small_model": "openai/gpt-5-nano"` to your `atlas.json` configuration file. 4. Ensure your main `"model"` setting in `atlas.json` points to a stronger tier model. 5. Utilize GPT-5 Nano as the summarizer for long Atlas sessions to minimize compaction costs. 6. Do not route the build agent to `openai/gpt-5-nano`; always check the TUI status line to confirm which slot it is filling. ## FAQ ### What is GPT-5 Nano's primary role within Atlas? GPT-5 Nano is best suited for Atlas's `small_model` slot, handling tasks like generating session titles, creating summaries for long sessions, and efficiently fanning out work to subagents. Its low cost makes these operations highly economical. ### How much does it cost to use GPT-5 Nano with Atlas? GPT-5 Nano is priced at $0.05 per Mtok for input tokens and $0.40 per Mtok for output tokens. This makes it the cheapest model in the entire OpenAI registry, offering significant cost savings. ### What is the context window size for GPT-5 Nano? GPT-5 Nano features a substantial 400K token context window. This is remarkable for its price point and allows Atlas to process large amounts of code or session history for tasks like bulk file scanning. ### Can GPT-5 Nano handle complex, multi step refactors in Atlas? No, GPT-5 Nano is not recommended for complex, multi step refactors or as the main build agent. It tends to lose the thread on intricate tasks, and its output quality degrades significantly before reaching its 128K max output limit. ### How does GPT-5 Nano compare to other GPT-5 models in terms of cost? GPT-5 Nano is exceptionally cost-effective. It is 25x cheaper than the standard GPT-5 model and 5x cheaper than gpt-5-mini for input tokens, making it the most economical choice in the OpenAI lineup. ### Why should I use GPT-5 Nano instead of a stronger model for some Atlas tasks? You should use GPT-5 Nano for tasks where cost efficiency and a broad context window are paramount, but the highest level of reasoning is not required. It significantly reduces operational costs for summarization and subagent work, freeing up budget for stronger models where they are truly needed. ### How do I configure Atlas to use GPT-5 Nano? To configure Atlas, set your `OPENAI_API_KEY`, run `atlas models openai`, and then add `"small_model": "openai/gpt-5-nano"` to your `atlas.json` file. Ensure your main `"model"` setting points to a more powerful model. ### What is the maximum output token limit for GPT-5 Nano? GPT-5 Nano has a documented maximum output of 128K tokens. However, its quality degrades long before reaching this limit, so it is best used for more concise outputs. --- Canonical HTML: https://seashell.sh/resources/models/gpt-5-nano Source of truth: aeo_pages row `/resources/models/gpt-5-nano` (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.