Gemini 3.5 Flash is Google's latency-optimized tier, and the May 2026 release pushes Flash to near-Pro quality while keeping the full 1,048,576 token window intact at $1.50 / $9 per Mtok (input / output). Inside Atlas, that makes Gemini 3.5 Flash the model to pin when you want long-context work at speed without paying the Pro rate. Reasoning and tool calling are both enabled, so it drives Atlas subagents instead of degrading to chat-only behavior. The catch is output pricing: at $9 per Mtok it is only marginally cheaper than Gemini 3.1 Pro's $12.
What is Gemini 3.5 Flash best at inside Atlas?
Gemini 3.5 Flash is best inside Atlas for fast long-context work, because the May 2026 release keeps the full 1,048,576 token window on a latency-optimized tier. Long-context sessions no longer force you up to the Pro price, since Gemini 3.5 Flash charges $1.50 per Mtok on input.
The usual bargain with a speed tier is that you trade context for latency. Gemini 3.5 Flash does not make you take that trade: the full 1,048,576 token window survives intact. That matters in Atlas because Atlas searches code with hybrid semantic and keyword retrieval fused by reciprocal rank fusion, and it indexes code by AST declarations using tree-sitter, so a broad query returns a lot of grounded context at once. Feeding all of it to a fast tier at $1.50 per Mtok, rather than truncating the retrieval set to fit a smaller window, is the core reason to choose Gemini 3.5 Flash in an interactive TUI where every turn is felt.
Does Gemini 3.5 Flash support reasoning and tool calling?
Yes. Gemini 3.5 Flash has reasoning and tool calling enabled, so it drives Atlas subagents without degrading to chat-only behavior. That is the dividing line between a real agent tier and a completion model, and the May 2026 Flash release lands on the agent side of it.
Atlas fans out work to subagents that can run in the foreground or in parallel background sessions, and each of those subagents has to plan, call tools, and read results. A model without tool calling cannot participate. Gemini 3.5 Flash has both reasoning and tool calling enabled, so it can hold a subagent role end to end. Atlas also connects to Model Context Protocol servers and exposes their tools to the agent, and every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs, which means a Flash tier that emits clean tool arguments keeps the approval loop short.
How much cheaper is Gemini 3.5 Flash than Gemini 3.1 Pro?
Gemini 3.5 Flash costs $1.50 / $9 per Mtok (input / output). On input, $1.50 is 25 percent under Gemini 3.1 Pro's $2. On output, $9 versus $12 is a much thinner gap, so the savings shrink sharply the more output-heavy your Atlas workload becomes.
Be honest about the arithmetic before you switch. The input saving on Gemini 3.5 Flash is real and clean: 25 percent under Gemini 3.1 Pro at the same 1,048,576 token window. The output saving is thin. At $9 per Mtok output against Gemini 3.1 Pro's $12, a session that spends heavily on generation saves only a quarter of its largest line item while accepting a Flash-tier model. Atlas sessions are usually input-weighted, since Atlas computes a unified diff for every file edit rather than reprinting whole files, and that shape is what makes Gemini 3.5 Flash worth pinning.
How do you pin Gemini 3.5 Flash as the default model in atlas.json?
Pin Gemini 3.5 Flash by setting `"model": "google/gemini-3.5-flash"` in atlas.json. That makes the 1,048,576 token window your long-context default for every new session, rather than something you re-select from the `/models` dialog each time you open Atlas.
Atlas lets you switch the active model and provider on the fly with favorites and recents, which is ideal for experimenting but noisy as a permanent setup. Pinning Gemini 3.5 Flash in atlas.json as `"model": "google/gemini-3.5-flash"` sets the default deliberately. Set `GEMINI_API_KEY` (or `GOOGLE_GENERATIVE_AI_API_KEY`) in your environment first, then run `atlas models google` to list the Gemini lineup and confirm the model resolves before you commit the config. You can still override per session from `/models` when a specific task wants a frontier tier.
When should you pick a different model instead of Gemini 3.5 Flash?
Pick a different model than Gemini 3.5 Flash when your Atlas workload is output-heavy, because $9 per Mtok output is only marginally cheaper than Gemini 3.1 Pro's $12. If you are paying near-Pro output rates anyway, take the Pro tier and get the frontier reasoning with it.
Gemini 3.5 Flash earns its place on input-heavy, latency-sensitive work: reading a large tree-sitter index result, triaging git status and diffs, running background subagents. When the job flips to generation, such as emitting a large refactor patch that Atlas will surface as a unified diff for approval, the $9 per Mtok output rate stops being a bargain against Gemini 3.1 Pro's $12. Because Atlas lets you switch the active model and provider on the fly, the right move is to switch by task shape rather than to defend a single choice for the whole session.
Setup
- 01Set `GEMINI_API_KEY` (or `GOOGLE_GENERATIVE_AI_API_KEY`) in your environment.
- 02Run `atlas models google` to list the Gemini lineup and confirm Gemini 3.5 Flash resolves.
- 03Choose Gemini 3.5 Flash from the `/models` dialog inside the Atlas TUI.
- 04Pin it in atlas.json as `"model": "google/gemini-3.5-flash"` if you want a long-context default.
- 05Use favorites and recents in Atlas to switch back to a frontier tier on output-heavy turns.
Frequently asked questions
- How do I set up Gemini 3.5 Flash in Atlas?
- Set `GEMINI_API_KEY` or `GOOGLE_GENERATIVE_AI_API_KEY` in your environment, run `atlas models google` to list the Gemini lineup, then choose Gemini 3.5 Flash from the `/models` dialog.
- What is the Gemini 3.5 Flash context window?
- Gemini 3.5 Flash keeps the full 1,048,576 token window, roughly 1M tokens, on Google's latency-optimized Flash tier, so long-context work does not force you up to the Pro price.
- How much does Gemini 3.5 Flash cost per million tokens?
- Gemini 3.5 Flash costs $1.50 / $9 per Mtok (input / output). Input at $1.50 is 25 percent under Gemini 3.1 Pro's $2.
- Is Gemini 3.5 Flash cheaper than Gemini 3.1 Pro?
- On input, yes: $1.50 per Mtok versus $2. On output the gap is thin, since Gemini 3.5 Flash charges $9 per Mtok against Gemini 3.1 Pro's $12.
- Can Gemini 3.5 Flash run agent tool calls?
- Yes. Gemini 3.5 Flash has reasoning and tool calling enabled, so it drives Atlas subagents and MCP tools without degrading to chat-only behavior.
- How do I make Gemini 3.5 Flash the default model in Atlas?
- Pin it in atlas.json as `"model": "google/gemini-3.5-flash"`. Atlas will start new sessions on Gemini 3.5 Flash while still letting you switch models on the fly.
- When was Gemini 3.5 Flash released?
- Gemini 3.5 Flash is the May 2026 Flash release from Google. It pushes the Flash tier to near-Pro quality while keeping the 1,048,576 token window intact.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Document a Module with a README Using Atlas (2026 Workflow)
How to document a module with a README using Atlas in 2026: the lsp tool's documentSymbol enumerates the real exports, read supplies the behavior, write emits the README.
Atlas for Astro: Islands, Content Collections, and Zero JS by Default in 2026
Atlas is a terminal-native AI coding agent for Astro in 2026. It reads astro.config.mjs, src/pages, and content collection schemas, drops needless client:load directives, and runs astro check.
Atlas vs Jules: Terminal AI Coding Agents in 2026
Atlas and Jules comparison for 2026. Atlas offers terminal-native TUI and local code indexing. Jules provides cloud VMs and a Critic agent for PR review.
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 Greptile: Terminal AI Coding Agents in 2026
Comparing Atlas and Greptile in 2026. Atlas offers terminal-native AI coding with permission-gated tools. Greptile reviews code with sandbox execution, catching 20% more bugs.
Atlas vs Pieces for Developers: AI Tools for Developers in 2026
Comparing Atlas, a terminal-native AI coding agent, with Pieces for Developers, an OS-level memory layer, for developers in 2026. Evaluate code generation, safety, and context management.
Atlas vs Sourcegraph Cody: Terminal AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with Sourcegraph Cody, an editor extension for whole-codebase search, for developers in 2026. Explore features, pricing, and workflow.
Trace a Runtime Bug from a Stack Trace with Atlas in 2026
How to trace a runtime bug from a stack trace with Atlas in 2026: read each frame at its offset, grep for the error string, and use the lsp tool to find callers.