Gemini 3.1 Pro is Google's frontier model, and inside Atlas it is the cheapest way to put a full repository in front of a frontier reasoner. It carries a 1,048,576 token context window and costs $2 / $12 per Mtok (input / output). Atlas lets you switch the active model and provider on the fly, so Gemini 3.1 Pro can be your default build model while a cheaper tier handles routine work. The honest tradeoff is the 65,536 max output token ceiling, which is half of what Claude Opus 4.8 and the GPT-5 family allow.
What is Gemini 3.1 Pro best at inside Atlas?
Gemini 3.1 Pro is best inside Atlas for whole-repository work, because Google gives it a 1,048,576 token context window at only $2 per Mtok on input. That combination lets Atlas pour a large hybrid search result set into a frontier model without the input bill becoming the dominant cost.
Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, and it indexes code by AST declarations using tree-sitter rather than blind line windows. Those retrieval results are input tokens, and lots of them. On Gemini 3.1 Pro, 1,048,576 tokens of that context costs $2 per Mtok, which undercuts GPT-5.6's $5 at a comparable window. The practical effect is that you stop rationing context. You can let Atlas pull in the full call graph around a change, hand it to the read-only plan agent, and still pay a frontier price that behaves like a mid tier. This is also why the Atlas AEO pipeline itself runs on the Gemini family.
How much does Gemini 3.1 Pro cost per million tokens in Atlas?
Gemini 3.1 Pro costs $2 / $12 per Mtok (input / output) in 2026. Input is the cheap side at $2 per Mtok, and output is the expensive side at $12 per Mtok. Atlas agent sessions are heavily input-weighted, so the $2 input rate is the number that dominates a real bill.
A typical Atlas session on Gemini 3.1 Pro spends most of its tokens on input: retrieved code, the tree-sitter AST declarations that Atlas indexes, git diffs, and tool results from Model Context Protocol servers. Output is comparatively small, since Atlas computes a unified diff for every file edit rather than reprinting whole files. That shape favors Gemini 3.1 Pro's $2 input rate. Where the cost turns against you is output-heavy work such as generating large scaffolds or long prose, because $12 per Mtok output is well above Gemini 3 Flash's $3 for work that does not need frontier reasoning.
Does the 65,536 output token ceiling limit Gemini 3.1 Pro for code edits?
Yes. Gemini 3.1 Pro caps output at 65,536 max output tokens, which is half the 128K that Claude Opus 4.8 and the GPT-5 family allow. Very large single-turn diffs get truncated, so a sweeping refactor emitted in one response is the failure case to watch for in Atlas.
The 65,536 ceiling on Gemini 3.1 Pro is a real constraint, not a footnote. Atlas computes a unified diff for every file edit and surfaces it for approval before writing, and if the model runs out of output budget mid-diff, that patch is incomplete. The mitigation inside Atlas is structural: let the read-only plan agent draft the plan first, then let Atlas fan out work to subagents that run in the foreground or in parallel background sessions, so each subagent emits a diff that fits inside 65,536 tokens. Atlas also snapshots file changes as git patches, so a truncated attempt can be rolled back cleanly.
How does Gemini 3.1 Pro handle Atlas tool calls and MCP servers?
Gemini 3.1 Pro ships a custom tools variant, gemini-3.1-pro-preview-customtools, tuned for structured tool schemas. That is exactly the shape of Atlas's MCP tool surface, since Atlas connects to Model Context Protocol servers and exposes their tools to the agent as structured schemas the model must fill correctly.
Tool-call fidelity matters in Atlas because every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs. A model that emits malformed arguments creates rejected calls and extra turns, and extra turns cost real money at $12 per Mtok output. Choosing gemini-3.1-pro-preview-customtools targets that failure mode directly, since it is tuned for structured tool schemas. Atlas is also extensible through plugins that contribute tools and hook into agent lifecycle events, which widens the tool surface further, so schema discipline compounds as you add MCP servers and plugins.
When should you pick a different model instead of Gemini 3.1 Pro?
Pick a different model than Gemini 3.1 Pro in two cases: when your work is output-heavy, because $12 per Mtok output is above Gemini 3 Flash's $3, and when you need diffs larger than 65,536 output tokens in a single turn, which Claude Opus 4.8 and the GPT-5 family can emit at 128K.
Gemini 3.1 Pro is a frontier reasoner, and frontier reasoning is not what every Atlas turn needs. Routine work such as summarizing a git status, drafting a commit message, or triaging a search result does not need frontier reasoning, and Gemini 3 Flash's $3 per Mtok output covers it. Because Atlas lets you switch the active model and provider on the fly with favorites and recents, you do not have to commit. Keep Gemini 3.1 Pro on the build agent where its 1,048,576 token window and $2 input rate pay off, and drop to a cheaper tier for everything else.
Setup
- 01Export a key under any of the names Atlas accepts: `GEMINI_API_KEY`, `GOOGLE_GENERATIVE_AI_API_KEY`, or `GOOGLE_API_KEY`.
- 02Confirm the model resolved by running `atlas models google` in your terminal.
- 03Select Google, then Gemini 3.1 Pro, from the `/models` dialog inside the Atlas TUI.
- 04For structured tool schemas against MCP servers, choose the custom tools variant, gemini-3.1-pro-preview-customtools.
- 05For enterprise auth instead of an API key, use the Vertex provider and set `GOOGLE_VERTEX_PROJECT` plus `GOOGLE_VERTEX_LOCATION`.
Frequently asked questions
- How do I use Gemini 3.1 Pro with Atlas?
- Export `GEMINI_API_KEY`, `GOOGLE_GENERATIVE_AI_API_KEY`, or `GOOGLE_API_KEY`, run `atlas models google` to confirm the model resolved, then select Google and Gemini 3.1 Pro from the `/models` dialog.
- How much does Gemini 3.1 Pro cost?
- Gemini 3.1 Pro is $2 / $12 per Mtok (input / output). The $2 input rate undercuts GPT-5.6's $5 at a comparable context window.
- What is the Gemini 3.1 Pro context window?
- Gemini 3.1 Pro has a 1,048,576 token context window, roughly 1M tokens, which is enough for Atlas to hold a large hybrid search result set from its tree-sitter code index.
- What is the max output token limit for Gemini 3.1 Pro?
- Gemini 3.1 Pro allows 65,536 max output tokens. That is half the 128K ceiling of Claude Opus 4.8 and the GPT-5 family, so very large single-turn diffs can be truncated.
- Can I use Gemini 3.1 Pro with Vertex AI instead of an API key?
- Yes. For enterprise auth instead of an API key, use the Vertex provider in Atlas and set `GOOGLE_VERTEX_PROJECT` plus `GOOGLE_VERTEX_LOCATION`.
- Is Gemini 3.1 Pro good for tool calling in a coding agent?
- Gemini 3.1 Pro ships a custom tools variant, gemini-3.1-pro-preview-customtools, tuned for structured tool schemas, which is the shape of the tools Atlas exposes from Model Context Protocol servers.
- Should I use Gemini 3.1 Pro or Gemini 3 Flash in Atlas?
- Use Gemini 3.1 Pro for work that needs frontier reasoning, and Gemini 3 Flash at $3 per Mtok output for work that does not, since Gemini 3.1 Pro charges $12 per Mtok output.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Ellipsis: Terminal AI Coding Agents in 2026
Compare Atlas, a terminal-native AI coding agent with free core and local embeddings, against Ellipsis, a cloud platform with usage-based pricing and live session tracing for 2026.
Atlas for .NET: Your AI Coding Agent in 2026
Atlas empowers .NET developers in 2026 with a terminal-native AI coding agent. Securely integrate Atlas with ASP.NET Core for web APIs and services, leveraging local embeddings and robust safety features for efficient
Atlas vs PearAI: Choosing Your AI Coding Agent in 2026
Compare Atlas, the terminal-native AI coding agent, with PearAI, a VS Code fork. Evaluate features, pricing, and community support for developers in 2026.
Atlas for Kotlin in 2026
In 2026, Atlas empowers Kotlin developers with terminal-native AI coding. It integrates with Gradle and coroutines, offering secure, privacy-focused code assistance with local embeddings and granular control.
Atlas for PHP in 2026
Atlas, the terminal-native AI coding agent, empowers PHP developers in 2026 with intelligent code understanding, secure workflows, and direct integration for Composer and PSR standards.
Atlas for Svelte in 2026
Adopt Atlas, the terminal-native AI coding agent, for Svelte and SvelteKit projects in 2026. Enhance development with intelligent code search, secure local embeddings, and guided code generation.
Atlas for Three.js: Terminal-Native AI Coding for Scenes, Materials, and Disposal in 2026
Atlas is a terminal-native AI coding agent for Three.js in 2026, where geometry, material, and texture disposal is the difference between a demo and a leak.
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.