Gemini 3.1 Pro Custom Tools is the model to drive Atlas with when your session is tool-dense. It is a distinct Google registry entry, gemini-3.1-pro-preview-customtools, that ships alongside standard Gemini 3.1 Pro as a tool-calling variant of the same checkpoint. It carries the same 1,048,576 token context, the same 65,536 token output ceiling, and the same $2 per Mtok input and $12 per Mtok output pricing, so picking the tool variant costs nothing extra. In Atlas every edit, read, grep, and bash call is a tool invocation, which is exactly the surface this checkpoint was built for.
What is Gemini 3.1 Pro Custom Tools in Atlas?
Gemini 3.1 Pro Custom Tools is a distinct Google registry entry, gemini-3.1-pro-preview-customtools, shipped alongside standard Gemini 3.1 Pro as a tool-calling variant of the same checkpoint. Inside Atlas it carries a 1,048,576 token context, a 65,536 token output ceiling, and $2 per Mtok input pricing.
Atlas is the terminal-native AI coding agent, and it lets you switch the active model and provider on the fly with favorites and recents. Gemini 3.1 Pro Custom Tools slots into that switcher as its own id, not as a flag on the base model. The practical meaning is that Google is exposing two paths through the same weights: gemini-3.1-pro-preview for general work, and gemini-3.1-pro-preview-customtools when the model is expected to spend most of its turns calling tools rather than writing prose. Because the two ids share the 1,048,576 token context and the 65,536 token output cap, there is no capability sacrifice in taking the tool path.
Why does a custom tools variant matter for a terminal coding agent?
Every action Atlas takes is a tool invocation: reading a file, running grep, executing bash, applying an edit. Gemini 3.1 Pro Custom Tools is purpose-built for that surface, which is why gemini-3.1-pro-preview-customtools exists as a separate id from gemini-3.1-pro-preview in Google's 2026 registry.
Consider what a single Atlas turn actually looks like. Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, indexes code by AST declarations using tree-sitter rather than blind line windows, then computes a unified diff for every file edit and surfaces it for approval before writing. Each of those steps is a tool call the model has to sequence correctly. On top of that, Atlas connects to Model Context Protocol servers and exposes their tools to the agent, and Atlas is extensible through plugins that contribute tools and hook into agent lifecycle events. A working Atlas session can therefore present a dense custom tool surface, and gemini-3.1-pro-preview-customtools is the id to reach for when it does.
How much does Gemini 3.1 Pro Custom Tools cost in Atlas?
Gemini 3.1 Pro Custom Tools costs $2 per Mtok input and $12 per Mtok output, identical to base Gemini 3.1 Pro. In 2026 that means the tool-calling variant is free to choose: the capability decision and the cost decision are cleanly separated, with no premium for the tool path.
The $2 per Mtok input figure is what governs most Atlas spend, because an agent reads far more than it writes. A large plan-then-build session pulls files, search results, and diffs into context repeatedly, and the 1,048,576 token window means Gemini 3.1 Pro Custom Tools can hold a lot of that at once. Output at $12 per Mtok is the expensive side, so the 65,536 token output ceiling is a budget lever as much as a capability one. If your Atlas sessions are heavy on reading and light on generation, Gemini 3.1 Pro Custom Tools is priced sensibly. If you generate enormous diffs, watch the output line.
What are the tradeoffs of gemini-3.1-pro-preview-customtools?
Gemini 3.1 Pro Custom Tools has two honest risks. Two near identical ids sit in the registry, and picking gemini-3.1-pro-preview instead of gemini-3.1-pro-preview-customtools silently changes tool behavior. Preview status also means Google may fold the variant back into the base model at any time.
The id confusion is the one that bites first. Because gemini-3.1-pro-preview and gemini-3.1-pro-preview-customtools differ only by a suffix, a typo in atlas.json produces a session that runs fine but sequences tool calls differently, with no error to tell you. Run `atlas models google` and read the id character by character. The preview risk is slower moving but real: a preview id is not a contract, and Google can retire or merge gemini-3.1-pro-preview-customtools. If you need a checkpoint that will not shift under a pinned atlas.json, a stable general availability id is the safer pin.
How do Atlas permissions work with Gemini 3.1 Pro Custom Tools?
Every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, and that gate applies identically to Gemini 3.1 Pro Custom Tools. Keeping Atlas in ask mode for the first few sessions on gemini-3.1-pro-preview-customtools lets you watch how the tool variant sequences reads, greps, and writes.
Permission gating is what makes running a preview tool-calling checkpoint reasonable in the first place. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, so a model that sequences tools aggressively still cannot write until you say so. Atlas computes a unified diff for every file edit and surfaces it for approval before writing, giving you a second checkpoint. Atlas also snapshots file changes as git patches so edits can be diffed and rolled back. With allow, ask, and deny rules set conservatively, the cost of an unexpected tool sequence from Gemini 3.1 Pro Custom Tools is a rejected diff, not a damaged working tree.
When should you pick a different model than Gemini 3.1 Pro Custom Tools?
Pick a different model than Gemini 3.1 Pro Custom Tools when you need a stable, non-preview id to pin in atlas.json for months, or when the session is not tool-dense enough to justify the variant. Atlas lets you switch the active model and provider on the fly, so this is a per-task call.
Gemini 3.1 Pro Custom Tools is a preview id. If reproducibility across weeks matters more than tool sequencing quality, a stable checkpoint is the better pin, and $2 per Mtok input plus $12 per Mtok output is not a bargain tier either. For high-volume background work, Atlas's small_model slot exists precisely so that titles, summaries, and subagent calls do not run on a frontier price. Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and those fan-outs are where a cheaper model pays for itself. Use Gemini 3.1 Pro Custom Tools where the tool surface is dense and the reasoning matters.
Setup
- 01Export GEMINI_API_KEY, or store the key via `atlas login` under the Google provider.
- 02Run `atlas models google` and look for gemini-3.1-pro-preview-customtools, which is distinct from gemini-3.1-pro-preview.
- 03Pin "model": "google/gemini-3.1-pro-preview-customtools" in atlas.json.
- 04Keep Atlas permissions in ask mode at first, using the allow, ask, and deny rules, so you can watch how the tool variant sequences calls.
- 05Review the unified diff Atlas shows before each write to sanity check the tool path.
Frequently asked questions
- what is gemini-3.1-pro-preview-customtools
- gemini-3.1-pro-preview-customtools is a distinct Google registry entry that ships alongside standard Gemini 3.1 Pro as a tool-calling variant of the same checkpoint. It has a 1,048,576 token context, a 65,536 token output ceiling, and costs $2 per Mtok input.
- how much does gemini 3.1 pro custom tools cost
- Gemini 3.1 Pro Custom Tools costs $2 per Mtok input and $12 per Mtok output, the same price as base Gemini 3.1 Pro. Choosing the custom tools variant carries no price premium.
- gemini 3.1 pro custom tools vs gemini 3.1 pro which should i use in atlas
- Use gemini-3.1-pro-preview-customtools when your Atlas session runs a dense custom tool surface, since every edit, read, grep, and bash call in Atlas is a tool invocation. Both ids share the same 1,048,576 token context and the same $2 per Mtok input price.
- how do i set gemini 3.1 pro custom tools as the model in atlas.json
- Set "model": "google/gemini-3.1-pro-preview-customtools" in atlas.json after exporting GEMINI_API_KEY and confirming the id with `atlas models google`. Check the suffix carefully, since gemini-3.1-pro-preview is a different id.
- what is the context window of gemini 3.1 pro custom tools
- Gemini 3.1 Pro Custom Tools has a 1,048,576 token context window and a 65,536 token output ceiling, identical to the base 3.1 Pro checkpoint, so nothing is traded away for the tool path.
- is gemini 3.1 pro custom tools safe to pin in production
- Gemini 3.1 Pro Custom Tools is a preview id, so Google may fold the variant back into the base model at any time. If you need a checkpoint that will not shift under a pinned atlas.json, choose a stable general availability id instead.
- does atlas let me switch models mid session
- Yes. Atlas lets you switch the active model and provider on the fly with favorites and recents, so you can move between gemini-3.1-pro-preview-customtools and another checkpoint without restarting the session.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for Bash in 2026
Atlas is a terminal-native AI coding agent for Bash in 2026, with shellcheck-aware suggestions. Harden a script or fix shellcheck warnings and review every diff.
Atlas vs Mistral Vibe for Code: Terminal AI Coding Agents in 2026
Compare Atlas and Mistral Vibe for Code in 2026. Atlas offers terminal-native TUI, explicit diffs, and BYO models. Mistral Vibe for Code provides a four-model stack, multi-platform access, and EU data sovereignty.
Atlas for Bun: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for Bun. Swap Node compatibility shims for Bun.serve and bun:sqlite, iterate with bun test, and format with bun fmt in 2026.
Atlas for Express in 2026
Atlas is a terminal-native AI coding agent for Express in 2026. It reads your middleware order, wraps async route handlers, and runs mocha behind a prompt.
Atlas vs OpenAI Codex CLI: Terminal AI Coding Agents in 2026
Comparing Atlas and OpenAI Codex CLI for terminal AI coding in 2026. Atlas offers a TUI, permission-gated tools, and BYO models, while OpenAI Codex CLI provides sandboxed execution and OpenAI model integration.
Atlas for Next.js in 2026
Adopt Atlas, the terminal-native AI coding agent, for Next.js development in 2026. Enhance productivity across App Router, server components, and API routes with secure, reviewable AI assistance.
Atlas for Groovy: A Terminal-Native AI Coding Agent for Gradle, Spock, and Jenkins in 2026
Atlas is a terminal-native AI coding agent for Groovy in 2026. It reads build.gradle closures and Jenkinsfiles, writes Spock specs, runs ./gradlew test, and applies Spotless.
Atlas for Actix Web in 2026
Atlas is a terminal-native AI coding agent for Actix Web in 2026. It reads extractors and app_data, then runs cargo test and cargo clippy behind a prompt.