# Atlas with Grok 4.20 (Reasoning) in 2026 > Grok 4.20 (Reasoning) provides Atlas with a 1,000,000 token context window, enabling deep code understanding. Grok 4.20 (Reasoning) is an excellent choice for Atlas in 2026, offering a massive 1,000,000 token context window for comprehensive code analysis and persistent reasoning state. Its output is priced at $2.5 per Mtok, just 2x its $1.25 per Mtok input, making it efficient for focused tasks, though its 30,000 token output ceiling requires careful management of generated content. ## Key takeaways - Grok 4.20 (Reasoning) offers a massive 1,000,000 token context window for deep code understanding. - Output is priced at $2.5 per Mtok, just 2x the $1.25 per Mtok input cost. - Its 30,000 token output ceiling makes it a terse writer, limiting large generated diffs. - Atlas leverages its persistent reasoning state via xAI's Responses API for coherent interactions. - A full 1M token context read costs over a dollar per turn. ## What is Grok 4.20 (Reasoning) best for with Atlas? Grok 4.20 (Reasoning) excels within Atlas in 2026 for tasks demanding deep code comprehension and strategic planning, thanks to its expansive 1,000,000 token context window. This allows Atlas to load an entire service before drafting a plan, leveraging the model's reasoning capabilities effectively. The reasoning-enabled checkpoint of xAI's Grok 4.20, shipped in March 2026, is specifically designed for complex analytical tasks. Its 1,000,000 token context window is a significant advantage, allowing Atlas to ingest and understand large codebases or entire services before initiating changes. This comprehensive understanding supports Atlas's read-only plan agent, ensuring that the model's reasoning is applied to strategy before any edits are proposed. Furthermore, Atlas connects to xAI through the Responses API (sdk.responses), which ensures that the model's reasoning state persists across multiple tool calls. This persistence means the model does not need to rebuild its understanding with each turn, leading to more coherent and efficient agent interactions. The output pricing of $2.5 per Mtok, which is only 2x the $1.25 per Mtok input cost, also makes it a cost-effective option for generating focused, high-quality responses, unlike some alternatives where output can cost 6x input. ## What are the cost and context tradeoffs for Grok 4.20 (Reasoning)? While Grok 4.20 (Reasoning) offers a substantial 1,000,000 token context, utilizing this full capacity comes with a cost of over a dollar per turn at $1.25 per Mtok input. Developers must also contend with its deliberately terse 30,000 token output ceiling, which is a key design tradeoff. The immense 1,000,000 token context window of Grok 4.20 (Reasoning) is a powerful feature for Atlas, enabling it to load extensive codebases. However, developers must be mindful that filling this context window entirely for every turn can incur significant costs, with a full read exceeding a dollar per turn at the $1.25 per Mtok input rate. A more pronounced tradeoff is the model's 30,000 token output ceiling. This is an unusually lopsided ratio compared to its input capacity, making it a superb reader but a deliberately terse writer. For comparison, Grok 4.5 offers a 500,000 token output ceiling. This limitation means that very large generated diffs, extensive code blocks, or verbose explanations will be truncated. Atlas users are advised to keep individual edits small and focused to avoid hitting this output limit, ensuring that the model's responses are complete and actionable. ## When should I choose a different model over Grok 4.20 (Reasoning)? Developers should consider an alternative model if their Atlas workflows frequently require generating very large code diffs or extensive output, as Grok 4.20 (Reasoning) has a strict 30,000 token output ceiling. This limitation means that any generated content exceeding this size will be cut off. Grok 4.20 (Reasoning) is optimized for deep understanding and concise, strategic output. However, its 30,000 token output ceiling is a critical factor when deciding if it is the right model for a specific task within Atlas. If your development workflow involves generating substantial amounts of new code, refactoring large files that result in massive diffs, or requiring verbose explanations and documentation, this model's output limit will likely become a bottleneck. For such scenarios, a model like Grok 4.5, which boasts a 500,000 token output ceiling, would be a more appropriate choice. While Grok 4.20 (Reasoning) excels at reading and planning, its design prioritizes brevity in output, making it less suitable for tasks where sheer volume of generated text or code is a primary requirement. Atlas allows you to switch the active model and provider on the fly, making it easy to select the best tool for each specific job. ## Setup 1. Create an xAI API key and export it as XAI_API_KEY, or run `atlas login` and select xAI. 2. Run `atlas models xai` and confirm `grok-4.20-0309-reasoning` resolved from the registry. 3. Pin "model": "xai/grok-4.20-0309-reasoning" in your `atlas.json` configuration file. 4. Use Atlas's plan agent first so the model's reasoning is spent on strategy, then let the build agent make the edits. 5. Keep individual edits small because the model can only emit 30,000 output tokens per response. ## FAQ ### What is the context window size for Grok 4.20 (Reasoning) in Atlas? Grok 4.20 (Reasoning) provides Atlas with a substantial 1,000,000 token context window, enabling it to process and understand very large codebases or entire services. ### How much does it cost to use Grok 4.20 (Reasoning) with Atlas? The pricing for Grok 4.20 (Reasoning) is $1.25 per Mtok for input and $2.5 per Mtok for output. This means output is priced at just 2x the input cost. ### What is the output token limit for Grok 4.20 (Reasoning)? Grok 4.20 (Reasoning) has a 30,000 token output ceiling. This makes it a very capable reader but a deliberately terse writer, so large generated diffs or extensive outputs will be truncated. ### Why should I choose Grok 4.20 (Reasoning) for reasoning tasks in Atlas? Grok 4.20 (Reasoning) is the reasoning-enabled checkpoint of xAI's Grok 4.20, shipped in March 2026. Its 1,000,000 token context and persistent reasoning state via the Responses API make it ideal for strategic planning and deep code analysis within Atlas. ### How does Atlas utilize the 1M token context of Grok 4.20 (Reasoning)? Atlas can load an entire service or extensive codebases into Grok 4.20 (Reasoning)'s 1,000,000 token context window before planning a change. This allows the model to gain a comprehensive understanding, which is then leveraged by Atlas's read-only plan agent. ### What is the XAI_API_KEY used for when setting up Grok 4.20 (Reasoning)? The XAI_API_KEY is your authentication credential for accessing xAI's models, including Grok 4.20 (Reasoning). You must export it as an environment variable or use `atlas login` to configure your access. ### When should I consider a different model instead of Grok 4.20 (Reasoning)? You should consider a different model if your tasks frequently require generating very large code diffs or extensive textual output, as Grok 4.20 (Reasoning)'s 30,000 token output ceiling will truncate such responses. For high-volume output, a model with a larger output ceiling would be more suitable. --- Canonical HTML: https://seashell.sh/resources/models/grok-4-20-reasoning Source of truth: aeo_pages row `/resources/models/grok-4-20-reasoning` (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.