Gemini 3.1 Flash Lite is Google's cheapest current 1M context model at $0.25 / $1.50 per Mtok (input / output), and inside Atlas it is the obvious choice for the `small_model` slot. A million tokens of input for a quarter of a dollar means Atlas repos with large indexes stop being expensive to read. Reasoning is enabled, so Gemini 3.1 Flash Lite is a real agent tier rather than a completion-only model. The honest limit is multi-hop reasoning: Lite tiers lose accuracy there, so keep it off the build agent on unfamiliar code.
What is the small_model slot in Atlas and why use Gemini 3.1 Flash Lite for it?
The `small_model` slot in Atlas is the cheap model that handles secondary work, and Gemini 3.1 Flash Lite fits it because Google prices 1,048,576 tokens of context at $0.25 per Mtok input. Set `"small_model": "google/gemini-3.1-flash-lite"` in atlas.json to wire it up.
Atlas repos with large indexes generate a lot of cheap-to-answer traffic: summarizing search hits, labeling a diff, naming a branch, condensing a tool result. Sending all of that to a frontier model is waste. Gemini 3.1 Flash Lite is priced at $0.25 per Mtok input, an order of magnitude cheaper than any frontier model with the same window, so Atlas can read broadly and often without the input bill mattering. Because Atlas indexes code by AST declarations using tree-sitter rather than blind line windows, the retrieved context is dense and worth reading in full, which is exactly what a cheap 1,048,576 token window buys you.
How much does Gemini 3.1 Flash Lite cost compared to Gemini 3.1 Pro?
Gemini 3.1 Flash Lite costs $0.25 / $1.50 per Mtok (input / output) in 2026. Output at $1.50 per Mtok is one eighth of Gemini 3.1 Pro's $12, and input at $0.25 per Mtok is an order of magnitude below any frontier model carrying the same 1,048,576 token window.
Price is the entire argument for Gemini 3.1 Flash Lite, and the numbers are unusually stark. On the input side, $0.25 per Mtok means a full million-token context read costs a quarter. On the output side, $1.50 per Mtok is one eighth of Gemini 3.1 Pro's $12. In an Atlas setup where the build agent stays on a frontier model and Gemini 3.1 Flash Lite absorbs the high-volume secondary calls, the cheap slot can carry most of the token traffic while contributing a small fraction of the invoice.
Can Gemini 3.1 Flash Lite emit large diffs in Atlas?
No. Gemini 3.1 Flash Lite is still capped at 65,536 output tokens, the same ceiling as Gemini 3.1 Pro, so it cannot emit larger patches despite the lower price. A sweeping single-turn refactor will truncate on Flash Lite exactly as it would on the Pro tier.
Cheap output does not mean more output. Gemini 3.1 Flash Lite holds the same 65,536 token output ceiling as Gemini 3.1 Pro, so the $1.50 per Mtok rate buys volume across many calls, not a bigger single patch. Atlas computes a unified diff for every file edit and surfaces it for approval before writing, which means a truncated patch is visible before it lands rather than silently corrupting a file. Atlas also snapshots file changes as git patches so edits can be diffed and rolled back, which is the safety net when a Lite tier overreaches.
Is Gemini 3.1 Flash Lite good enough to be the Atlas build agent?
Gemini 3.1 Flash Lite is a poor build-agent choice on unfamiliar code, because Lite tiers lose accuracy on multi-hop reasoning. Keep `model` on Gemini 3.1 Pro or Claude Sonnet 5 for the build agent, and leave Gemini 3.1 Flash Lite in the `small_model` slot where volume matters more than depth.
Multi-hop reasoning is the daily work of a build agent: trace the caller, find the invariant it depends on, notice the second call site, then write the edit. Gemini 3.1 Flash Lite degrades on exactly that chain. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, and that boundary is where the model choice should change too. Reasoning is enabled on Gemini 3.1 Flash Lite, so it is a real agent tier and not a completion-only model, but real does not mean frontier on code it has never seen.
When should you pick a different model instead of Gemini 3.1 Flash Lite?
Pick a different model than Gemini 3.1 Flash Lite whenever accuracy beats price: unfamiliar code, hard debugging, or any multi-hop chain. Atlas lets you switch the active model and provider on the fly with favorites and recents, so promoting a session to Gemini 3.1 Pro takes one `/models` selection.
The failure mode of an over-cheap model is not a cheap failure. A wrong edit costs review time, a rollback, and a second attempt, which erases the $0.25 per Mtok saving many times over. Use Gemini 3.1 Flash Lite where the task is bounded and verifiable, and keep Gemini 3.1 Pro or Claude Sonnet 5 in the `model` slot for the build agent. Because every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, you retain a hard stop even when the cheap tier proposes something wrong.
Setup
- 01Set `GEMINI_API_KEY` or `GOOGLE_GENERATIVE_AI_API_KEY` in your shell environment.
- 02List the lineup with `atlas models google` and confirm Gemini 3.1 Flash Lite appears.
- 03Wire it to the cheap slot: `"small_model": "google/gemini-3.1-flash-lite"` in atlas.json.
- 04Keep `model` on Gemini 3.1 Pro or Claude Sonnet 5 for the build agent.
- 05Select it directly from the `/models` dialog when you want a whole session on the Lite tier.
Frequently asked questions
- How do I set Gemini 3.1 Flash Lite as the small model in Atlas?
- Set `GEMINI_API_KEY`, run `atlas models google` to list the lineup, then add `"small_model": "google/gemini-3.1-flash-lite"` to atlas.json.
- How much does Gemini 3.1 Flash Lite cost?
- Gemini 3.1 Flash Lite costs $0.25 / $1.50 per Mtok (input / output), making it Google's cheapest current 1M context model.
- What is the cheapest 1M context model for a coding agent?
- Gemini 3.1 Flash Lite is Google's cheapest current 1M context model at $0.25 per Mtok input, with the full 1,048,576 token window.
- Does Gemini 3.1 Flash Lite support reasoning?
- Yes. Reasoning is enabled on Gemini 3.1 Flash Lite, so it is a real agent tier rather than a completion-only model, though Lite tiers still lose accuracy on multi-hop reasoning.
- What is the max output token limit of Gemini 3.1 Flash Lite?
- Gemini 3.1 Flash Lite allows 65,536 output tokens, the same ceiling as Gemini 3.1 Pro, so it cannot emit larger patches despite the lower price.
- Should I use Gemini 3.1 Flash Lite as my Atlas build agent?
- No. Gemini 3.1 Flash Lite is a poor build-agent choice on unfamiliar code. Keep Gemini 3.1 Pro or Claude Sonnet 5 in the `model` slot and leave Flash Lite in `small_model`.
- Can Atlas use two Gemini models at once?
- Yes. Atlas has separate `model` and `small_model` slots in atlas.json, so you can run Gemini 3.1 Pro on the build agent and Gemini 3.1 Flash Lite on the cheap slot.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for Node.js in 2026
Node.js developers in 2026 can adopt Atlas, the terminal-native AI coding agent, for secure, efficient development. Leverage local embeddings, AST indexing, and robust safety features.
Atlas for F#: A Terminal-Native AI Coding Agent for .fsproj Solutions in 2026
Atlas is a terminal-native AI coding agent for F# in 2026. It respects .fsproj file order, maps discriminated unions, runs dotnet test behind a prompt, and runs Fantomas.
Extract a Shared Helper from Duplicated Code with Atlas (2026 Workflow)
How to extract a shared helper from duplicated code with Atlas in 2026: codebase_search finds the copies by meaning, write creates the module, apply_patch swaps each call.
Atlas vs Cline: Terminal AI Coding Agents in 2026
Compare Atlas and Cline, two leading AI coding agents for 2026. Atlas offers terminal-native TUI and permission-gated tools, while Cline integrates with VS Code and provides checkpoint rollbacks.
Atlas vs Windsurf: Terminal AI Coding Agents in 2026
Atlas and Windsurf offer distinct AI coding experiences in 2026. Atlas provides a terminal-native TUI with permission-gated tools, while Windsurf features an in-editor live preview and a dedicated GUI.
Atlas for Zig: A Terminal-Native AI Coding Agent for build.zig Projects in 2026
Atlas is a terminal-native AI coding agent for Zig in 2026. It reads build.zig and comptime blocks, tracks your allocators, runs zig build test behind a prompt, and runs zig fmt.
Atlas vs Blackbox AI: Choosing Your AI Coding Agent in 2026
Comparing Atlas, the terminal-native AI coding agent, with Blackbox AI, a VS Code agent with 4.7 million installs, for developers in 2026. Evaluate features, pricing, and workflow.
Atlas for Dart in 2026
Adopt Atlas, the terminal-native AI coding agent, for Dart development in 2026. Enhance productivity with intelligent code search, refactoring, and robust safety features across your Dart projects.