GitHub Models offers a highly accessible entry point for driving Atlas, the terminal-native AI coding agent, in 2026. It provides a diverse catalog of models, including OpenAI o3, xAI Grok 3, and AI21 Jamba 1.5 Large, all at $0 per Mtok, making it an excellent choice for evaluation and prototyping within your existing GitHub ecosystem.
What are GitHub Models best for with Atlas?
GitHub Models are best suited for developers in 2026 looking to quickly integrate Atlas into their workflow without any upfront cost or complex setup. With all models listed at $0 per Mtok, it's ideal for initial evaluation and prototyping, leveraging your existing GITHUB_TOKEN for authentication.
For developers using Atlas, GitHub Models offers unparalleled ease of access. Authentication is streamlined, requiring only a plain GITHUB_TOKEN, the same token your `gh CLI` already uses. This eliminates the need for separate sign-ups, credit card details, or managing multiple API keys across different providers. Atlas, with its ability to switch the active model and provider on the fly, benefits greatly from this low-friction setup, allowing you to experiment with various models like OpenAI o3, xAI Grok 3, and DeepSeek-R1-0528 through a single endpoint. This makes GitHub Models an excellent choice for initial exploration of Atlas's capabilities, such as its hybrid semantic and keyword retrieval fused by reciprocal rank fusion (Axis), or its read-only plan agent that drafts plans before switching to a build agent. The inclusion of AI21 Jamba 1.5 Large, offering a 256,000 token context window for $0, provides a substantial capacity for complex code analysis and generation tasks during the prototyping phase.
What are the cost and context window tradeoffs for GitHub Models?
While GitHub Models offers an attractive $0 per Mtok pricing model, included with a GitHub account, it comes with specific tradeoffs in 2026, particularly regarding rate limits and context window sizes. The AI21 Jamba 1.5 Large model provides a generous 256,000 token context window, but other models like GPT-4.1 are capped at 128,000 tokens.
The primary advantage of GitHub Models is its cost: all models in its registry are listed at $0 per Mtok, functioning on a rate-limited rather than metered basis. This means you won't incur direct costs for usage. However, this free access is balanced by strict rate limits, which are tiered according to your GitHub plan. For a real agent loop within Atlas, especially one utilizing parallel subagents, these limits can be quickly encountered. Therefore, GitHub Models is explicitly positioned for evaluation and prototyping, not for production workloads where consistent, high-volume access is critical. Regarding context windows, GitHub Models provides 256,000 tokens for AI21 Jamba 1.5 Large and Mini, and 200,000 tokens for OpenAI o3, o1, and o4-mini. It's important to note that some context windows are trimmed compared to their first-party counterparts. For instance, GPT-4.1 is capped at 128,000 tokens here, significantly less than the 1,047,576 tokens available directly from OpenAI. Similarly, DeepSeek-R1 is limited to 65,536 tokens, whereas it offers 163,840 elsewhere. These trimmed windows mean that while the models are free, they may not handle the largest codebases or most extensive context requirements as effectively as direct provider access would.
When should I choose a different model for Atlas?
You should consider an alternative model for Atlas in 2026 if your project demands consistent, high-volume AI interactions or requires the full context window capabilities of models like GPT-4.1. GitHub Models' strict rate limits and trimmed context windows, such as GPT-4.1 being capped at 128,000 tokens, make it unsuitable for production-grade applications.
While GitHub Models is excellent for initial setup and experimentation with Atlas, it's not designed for every use case. If your Atlas workflow involves intensive, continuous operations, such as driving multiple subagents in parallel for complex refactoring or extensive code generation, you will likely hit the strict rate limits. These limits, tiered by your GitHub plan, prevent GitHub Models from being a viable option for production environments where uninterrupted service is paramount. Furthermore, if your work with Atlas requires the absolute largest context windows available from specific models, you will find GitHub Models' offerings to be trimmed. For example, if you need the full 1,047,576 token context of GPT-4.1, you would need to access it directly from OpenAI, as GitHub Models caps it at 128,000 tokens. Similarly, DeepSeek-R1's context is reduced from 163,840 to 65,536 tokens. For tasks that involve indexing extremely large codebases using Atlas's AST declarations with tree-sitter, or for comprehensive analysis that benefits from maximum context, a direct provider integration might be more appropriate. In such scenarios, the cost savings of GitHub Models would be outweighed by the performance limitations imposed by rate limits and reduced context windows.
How does GitHub Models simplify Atlas setup?
GitHub Models significantly simplifies the setup process for Atlas in 2026 by leveraging your existing GitHub authentication. You can get Atlas talking to powerful models like OpenAI o3 and AI21 Jamba 1.5 Large with just a few commands, using the same GITHUB_TOKEN you already have for your `gh CLI`.
The setup for GitHub Models with Atlas is designed for minimal friction. Instead of managing multiple API keys, billing accounts, and provider-specific configurations, you only need your GITHUB_TOKEN. This token, whether a classic Personal Access Token (PAT) or the output from `gh auth token`, simply needs to have models access enabled and be exported as an environment variable. Atlas then uses this single token to access a catalog spanning five different vendors, including OpenAI, xAI, DeepSeek, Cohere, and Meta. This unified access point means you can quickly switch between models like `github-models/openai/gpt-4.1` for reasoning tasks and `github-models/openai/gpt-4.1-mini` for smaller calls like title and summary generation, all within your `atlas.json` configuration. This approach not only saves time but also reduces the administrative overhead typically associated with integrating multiple AI providers into your development workflow.
Setup
- 01Export your GITHUB_TOKEN environment variable. This can be a classic Personal Access Token (PAT) or the token printed by `gh auth token`, ensuring it has models access enabled.
- 02Run `atlas models github-models` in your terminal to list the resolved catalog of available models. The base URL for inference is `https://models.github.ai/inference`.
- 03Pin your preferred model for reasoning in your `atlas.json` configuration. For example, set `"model": "github-models/openai/gpt-4.1"` or `"model": "github-models/openai/o3"`.
- 04To conserve your primary model's rate limit, configure a smaller model for less demanding tasks. Set `"small_model": "github-models/openai/gpt-4.1-mini"` in your `atlas.json` for title and summary calls.
Frequently asked questions
- What is the cost of using GitHub Models with Atlas?
- All models listed in the GitHub Models registry are available at $0 per Mtok. This access is included with a GitHub account and is rate-limited rather than metered, meaning there are no direct usage charges.
- What is the largest context window available with GitHub Models?
- GitHub Models provides AI21 Jamba 1.5 Large and Mini with a 256,000 token context window, which is the longest free context window available in the catalog.
- Can I use GitHub Models for production applications with Atlas?
- No, GitHub Models is intended for evaluation and prototyping with Atlas. Its strict rate limits, tiered by GitHub plan, make it unsuitable for production workloads that require consistent, high-volume AI interactions.
- How do I authenticate Atlas with GitHub Models?
- Authentication is straightforward: you simply export your GITHUB_TOKEN environment variable. This is the same token your `gh CLI` uses, and it must have models access enabled.
- Are the context windows for models like GPT-4.1 the same as direct access?
- No, context windows for some models are trimmed. For example, GPT-4.1 is capped at 128,000 tokens when accessed via GitHub Models, compared to 1,047,576 tokens available directly from OpenAI.
- Which AI vendors are available through GitHub Models?
- GitHub Models provides access to models from five vendors through a single endpoint: OpenAI (o3, o1, o4-mini), xAI (Grok 3), DeepSeek (DeepSeek-R1-0528), Cohere (Command A), Meta (Llama 4 Maverick FP8), and Microsoft (Phi-4).
- How does Atlas handle different models for different tasks with GitHub Models?
- Atlas allows you to configure a primary model for reasoning (e.g., `github-models/openai/o3`) and a `small_model` (e.g., `github-models/openai/gpt-4.1-mini`) for less intensive tasks like title and summary generation, helping to manage your rate limits effectively.
Try SeaShell in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install SeaShellRelated guides
Atlas vs Factory AI: Terminal AI Coding Agents in 2026
Compare Atlas and Factory AI in 2026. Atlas offers terminal-native TUI, permission-gated tools, and local embeddings. Factory AI provides Droids and cloud machines for enterprise.
Atlas for Haskell in 2026
Atlas is a terminal-native AI coding agent for Haskell in 2026. Run it in a project with a .cabal file or a stack.yaml, let cabal build drive it, and review the diff.
Atlas vs Aider: Terminal AI Coding Agents in 2026
Compare Atlas and Aider, two terminal AI coding agents for 2026. Atlas offers a rich TUI, interactive plan approval, and plugin support. Aider features git-revision commits and repo-map awareness.
Atlas vs OpenHands: Terminal AI Coding Agents in 2026
Comparing Atlas and OpenHands in 2026 for AI coding agents. Atlas offers terminal-native TUI and explicit change review, while OpenHands provides an open-source platform with skills and a limited free cloud tier.
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().
Add a Regression Test for a Bug Fix with Atlas in 2026
How to add a regression test with Atlas in 2026: red first, then green. bash records the exit code, write creates the failing test, and edit applies the fix.
Research a Third-Party API Before Integrating It with Atlas in 2026
How to research a third-party API with Atlas in 2026: websearch finds the current docs, webfetch pulls the page as markdown or text, and grep checks repo conventions.
Atlas for Deno: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for Deno. Tighten --allow-all down to real permission scopes, add JSR deps, and run deno test and deno check in 2026.