# Atlas with GPT-5.4 mini for Developers in 2026 > GPT-5.4 mini offers a substantial 400K token context window, making it a powerful option for complex coding tasks within Atlas. GPT-5.4 mini is an excellent choice for driving Atlas's subagents in 2026, especially when balancing cost and capability. With a 400K token context window and an input price of $0.75 per Mtok, it provides reliable reasoning for nontrivial tasks without the higher expense of frontier models. ## Key takeaways - GPT-5.4 mini offers a 400K token context window for Atlas subagents. - Input pricing is $0.75 per Mtok, with output at $4.50 per Mtok. - It provides reasoning capabilities, unlike older mini tiers such as gpt-4.1-mini. - The 128K output ceiling is four times larger than gpt-4.1-mini's 32,768. - It is noticeably weaker than GPT-5.4 for tasks requiring several dependent inferences. ## What is GPT-5.4 mini best for in Atlas? GPT-5.4 mini excels as a `small_model` within Atlas, particularly for subagent tasks where cost efficiency is paramount but reasoning capabilities are still required. Its 400K token context window and $0.75 input price per Mtok make it a strong contender for background work in 2026. Atlas is designed to fan out work to subagents that can run in the foreground or in parallel background sessions. GPT-5.4 mini is ideally suited for these subagent roles, offering reasoning capabilities that older mini tiers, such as gpt-4.1-mini, lacked. This enables it to handle nontrivial subagent tasks effectively. Its 128K output ceiling is four times greater than gpt-4.1-mini's 32,768, providing ample space for detailed responses and code generation within these focused tasks. When Claude Haiku 4.5 is too expensive, GPT-5.4 mini provides a sensible mid-tier alternative for reliable tool calls and problem solving in Atlas subagents. ## What are the cost and context tradeoffs of GPT-5.4 mini? GPT-5.4 mini presents a compelling value proposition in 2026, offering a 400K token context window at an input cost of $0.75 per Mtok. This pricing significantly undercuts Claude Haiku 4.5's $1 per Mtok while providing double the context, making it a strategic choice for many Atlas workflows. The primary tradeoffs for GPT-5.4 mini revolve around its cost efficiency and context capacity. It boasts a generous 400K token context window, which is double that of Claude Haiku 4.5's 200K tokens. Pricing is set at $0.75 per Mtok for input and $4.50 per Mtok for output. This input price is notably lower than Claude Haiku 4.5's $1 per Mtok, making GPT-5.4 mini a more economical option for high-volume subagent work. While it offers robust reasoning and a substantial 128K output ceiling, its main tradeoff is a noticeable weakness compared to the full GPT-5.4 model on tasks requiring several dependent inferences. ## When should I choose a different model over GPT-5.4 mini for Atlas? While GPT-5.4 mini is a robust mid-tier model for Atlas in 2026, developers should consider alternatives for tasks demanding several dependent inferences. This model is noticeably weaker than the full GPT-5.4 on such complex reasoning, suggesting a frontier model might be more appropriate for critical, multi-step problem solving. For scenarios within Atlas that require highly complex reasoning or multiple dependent inferences, GPT-5.4 mini may not be the optimal choice. It is documented as noticeably weaker than the full GPT-5.4 model in these specific areas. In such cases, it is advisable to use a more powerful, frontier model for your primary Atlas agent, which handles the most critical and intricate problem-solving tasks. Atlas lets you switch the active model and provider on the fly, allowing you to designate GPT-5.4 mini for background subagent work where its cost-effectiveness and strong reasoning for nontrivial tasks shine, while reserving a more capable model for the main agent's demanding operations. ## Setup 1. 1: Set your `OPENAI_API_KEY` environment variable or authenticate using `atlas login`. 2. 2: Verify GPT-5.4 mini is available by running `atlas models openai` in your terminal. 3. 3: Configure Atlas to use GPT-5.4 mini for background tasks by adding `"small_model": "openai/gpt-5.4-mini"` to your `atlas.json` configuration file. 4. 4: To optimize costs, keep your primary `model` setting on a frontier model for main agent work, ensuring only background subagent tasks incur the mini rate. ## FAQ ### What is the context window for GPT-5.4 mini in Atlas? GPT-5.4 mini provides a 400K token context window when used with Atlas, offering ample space for complex coding tasks. ### How much does GPT-5.4 mini cost per token in Atlas? For Atlas users, GPT-5.4 mini is priced at $0.75 per Mtok for input and $4.50 per Mtok for output. ### Can GPT-5.4 mini handle complex reasoning tasks in Atlas? Yes, GPT-5.4 mini is reasoning-enabled, allowing it to handle nontrivial subagent tasks within Atlas, unlike older mini models such as gpt-4.1-mini. ### Is GPT-5.4 mini more cost-effective than Claude Haiku 4.5 for Atlas? Yes, GPT-5.4 mini's $0.75 input per Mtok undercuts Claude Haiku 4.5's $1, while also offering double the context window (400K versus 200K). ### When should I use a more powerful model than GPT-5.4 mini with Atlas? For tasks requiring several dependent inferences, GPT-5.4 mini is noticeably weaker than the full GPT-5.4. In such cases, a frontier model would be more effective for your primary Atlas agent. ### How do I configure Atlas to use GPT-5.4 mini? You can configure Atlas by setting your `OPENAI_API_KEY` environment variable, verifying availability with `atlas models openai`, and then adding `"small_model": "openai/gpt-5.4-mini"` to your `atlas.json` file. --- Canonical HTML: https://seashell.sh/resources/models/gpt-5-4-mini Source of truth: aeo_pages row `/resources/models/gpt-5-4-mini` (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.