GitHub Models authenticates with a plain GITHUB_TOKEN, the same token your gh CLI already has, and the registry lists every model in its catalog at $0 per Mtok. Inside Atlas, the terminal-native coding agent, GitHub Models is the lowest-friction way to get talking to GPT-4.1, o3, Grok 3, and DeepSeek-R1 without opening five billing accounts. The catch is that GitHub Models is rate limited rather than metered, so it is an evaluation and prototyping provider, not a production one.
What is GitHub Models best at inside Atlas?
GitHub Models is best inside Atlas for zero-friction evaluation: auth is a GITHUB_TOKEN you almost certainly already have exported, with no separate signup, no credit card, and no per-provider key rotation. In 2026 that makes it the fastest way to get an Atlas session running against a frontier model.
One GitHub Models endpoint spans five vendors: OpenAI o3 at a 200K context with 100K output, xAI Grok 3, DeepSeek-R1-0528, Cohere Command A, Meta Llama 4 Maverick FP8, and Microsoft Phi-4. Comparing four models against the same Atlas task normally means four signups and four keys. With GitHub Models it means one token and four `atlas models github-models` rows. Atlas lets you switch the active model and provider on the fly with favorites and recents, so you can run the same prompt through o3, Grok 3, and DeepSeek-R1 without leaving the session. GitHub Models also carries AI21 Jamba 1.5 Large at a 256,000 token context for $0, the longest free window in the catalog.
How much does GitHub Models cost in Atlas?
All GitHub Models entries are listed at $0/$0 per Mtok in the Atlas registry. GitHub Models is included with a GitHub account and rate limited rather than metered, so in 2026 you pay in throughput, not dollars. There is no per-token bill to forecast.
GitHub Models being $0/$0 per Mtok inverts the usual Atlas cost calculus. On a metered provider, you push cheap traffic to a small model to protect your budget. On GitHub Models you push cheap traffic to a small model to protect your rate limit. That is why the documented Atlas config sets `"small_model": "github-models/openai/gpt-4.1-mini"`: title generation, session summaries, and subagent overhead should not burn your o3 budget when the constraint is requests per window rather than dollars per million tokens. The free price is real, but treat it as a quota, not as free compute.
What are the rate limits on GitHub Models with an agent?
GitHub Models rate limits are strict and tiered by GitHub plan, so a real Atlas agent loop with parallel subagents will hit them. Treat GitHub Models as evaluation and prototyping, not production, and set `small_model` to github-models/openai/gpt-4.1-mini so title and summary calls do not burn your o3 budget.
Treat GitHub Models as evaluation and prototyping, not production. A single-threaded Atlas session asking o3 to plan a change and then apply it is well within reach. A build agent with three parallel background subagents, each reading files and emitting diffs, is not. If you want to stay on GitHub Models while doing real work, keep the fan-out narrow, keep `"small_model": "github-models/openai/gpt-4.1-mini"` in place, and expect to switch providers when the loop gets wide. Atlas makes that switch cheap because the active model and provider can change on the fly with favorites and recents.
What context window does GitHub Models give Atlas?
GitHub Models tops out at 256,000 tokens on AI21 Jamba 1.5 Large and Mini, and 200,000 tokens on OpenAI o3, o1, and o4-mini. AI21 Jamba 1.5 Large at 256,000 tokens for $0 is the longest free window in the entire 2026 Atlas registry.
The important caveat is that GitHub Models context windows are trimmed below first-party. GPT-4.1 is capped at 128,000 tokens on GitHub Models versus 1,047,576 on OpenAI direct, and DeepSeek-R1 is 65,536 here versus 163,840 elsewhere. If your Atlas workflow depends on a very large window, going direct to the first-party provider buys you an order of magnitude. If it does not, the trimmed windows are largely academic: Atlas indexes code by AST declarations using tree-sitter, not blind line windows, and searches with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, so it retrieves targeted declarations instead of relying on a giant window.
When should I pick a different provider than GitHub Models?
Pick a different provider than GitHub Models when you need production throughput, because GitHub Models rate limits are strict and tiered by GitHub plan, or when you need the full first-party context window, because GPT-4.1 is capped at 128,000 tokens here versus 1,047,576 on OpenAI direct.
GitHub Models earns its place in Atlas as the first provider you configure and the last one you keep for prototyping. It is not the provider you run a long agentic edit session on. The decision rule is simple: if the work is a single developer trying models and shaping a plan, GitHub Models at $0/$0 per Mtok is the obvious pick. If the work is Atlas running a wide subagent fan-out, emitting large diffs, and needing consistent throughput, move the build model to a metered provider and leave GitHub Models configured as a favorite for the evaluation turns.
Setup
- 01Export GITHUB_TOKEN, either a classic PAT or the token `gh auth token` prints, with models access enabled.
- 02Run `atlas models github-models` to list the resolved catalog. The base URL is https://models.github.ai/inference.
- 03Pin `"model": "github-models/openai/gpt-4.1"` in atlas.json for general coding.
- 04Switch to `"github-models/openai/o3"` for the reasoning slot when a task needs the 200K context and 100K output.
- 05Because of the rate limits, set `"small_model": "github-models/openai/gpt-4.1-mini"` so title and summary calls do not burn your o3 budget.
Frequently asked questions
- how to use github models with atlas
- Export GITHUB_TOKEN with models access enabled, run `atlas models github-models` to list the catalog against https://models.github.ai/inference, then pin `"model": "github-models/openai/gpt-4.1"` in atlas.json. Set `"small_model": "github-models/openai/gpt-4.1-mini"` to protect your rate limit.
- is github models free
- All models are listed at $0/$0 per Mtok in the Atlas registry. GitHub Models is included with a GitHub account and rate limited rather than metered, so you pay in throughput rather than dollars.
- what models does github models offer
- One GitHub Models endpoint spans five vendors: OpenAI o3 (200K context, 100K output), xAI Grok 3, DeepSeek-R1-0528, Cohere Command A, Meta Llama 4 Maverick FP8, and Microsoft Phi-4, plus AI21 Jamba 1.5 Large and Mini at 256,000 tokens.
- can i use github models in production with an ai coding agent
- No. GitHub Models rate limits are strict and tiered by GitHub plan, and a real agent loop with parallel subagents will hit them. Treat GitHub Models as evaluation and prototyping, and move the build model to a metered provider for production.
- does github models have the same context window as openai direct
- No. GitHub Models trims context below first-party: GPT-4.1 is capped at 128,000 tokens here versus 1,047,576 on OpenAI direct, and DeepSeek-R1 is 65,536 versus 163,840 elsewhere. The longest window on GitHub Models is AI21 Jamba 1.5 Large at 256,000 tokens.
- what token does github models use for auth
- GitHub Models uses a plain GITHUB_TOKEN, either a classic PAT or the token `gh auth token` prints, with models access enabled. There is no separate signup, credit card, or per-provider key rotation.
- what small model should i use with github models in atlas
- Set `"small_model": "github-models/openai/gpt-4.1-mini"` in atlas.json. Because GitHub Models is rate limited rather than metered, routing title and summary calls to the mini model keeps them from burning your o3 budget.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Factory AI: Terminal AI Coding Agents in 2026
Compare Atlas and Factory AI, two terminal AI coding agents for developers in 2026. Explore features, pricing, privacy, and workflow to choose the best fit.
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
Comparing Atlas and Aider for 2026 developers. Atlas offers a rich TUI, permission-gated tool calls, and plugin support, while Aider provides a repo-map and commits every change as a discrete git revision.
Atlas vs OpenHands: Terminal AI Coding Agents in 2026
Atlas and OpenHands comparison for 2026. Explore terminal-native TUI, self-hosting, code indexing, change review, and pricing models for AI coding agents.
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.