Gemini 2.5 Pro is the June 2025 Gemini flagship and still a strong default for long-context coding work inside Atlas. It was the checkpoint that brought the 1,048,576 token window plus a 65,536 token output cap to the Pro tier, and at $1.25 per Mtok input and $10 per Mtok output it is cheaper to read with than Gemini 3 Pro. Its biggest practical advantage in 2026 is that gemini-2.5-pro is a stable GA id, not a preview, so behavior does not shift underneath a pinned atlas.json.
What is Gemini 2.5 Pro best at inside Atlas?
Gemini 2.5 Pro is best at long-context reading inside Atlas. Released as the June 2025 Gemini flagship, gemini-2.5-pro pairs a 1,048,576 token context with a $1.25 per Mtok input price, which makes ingesting a large repository before a refactor materially cheaper than on a newer Pro checkpoint.
Atlas is the terminal-native AI coding agent, and its first job on any nontrivial task is reading. 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, so what lands in context is dense and relevant. Feeding that into a 1,048,576 token window at $1.25 per Mtok is the core economic argument for Gemini 2.5 Pro. Reasoning is enabled, and the 65,536 token output cap sits comfortably above the shorter-output profile of older models, so 2.5 Pro can still emit a real diff after it has read.
How much cheaper is Gemini 2.5 Pro than Gemini 3 Pro?
Gemini 2.5 Pro costs $1.25 per Mtok input, which is 37 percent cheaper than Gemini 3 Pro's $2 for the same 1,048,576 token window. On output the gap narrows: $10 per Mtok on 2.5 Pro against $12 on Gemini 3 Pro is only a modest saving.
The asymmetry matters when you model a real Atlas session. Agent workloads are read-heavy, so the input price is the line that moves the bill, and that is where Gemini 2.5 Pro wins by 37 percent. Output is the smaller share, and $10 per Mtok output on gemini-2.5-pro is close enough to $12 that it should not drive the decision on its own. In short, Gemini 2.5 Pro is chosen for cheap reading, not for cheap writing. If your Atlas usage skews toward generating large diffs rather than exploring code, the saving thins out considerably.
Why does a stable GA id like gemini-2.5-pro matter?
gemini-2.5-pro is a stable GA id, not a preview, which means a pinned atlas.json keeps behaving the same way tomorrow. For teams benchmarking Atlas prompts or chasing a regression across 2026, that reproducibility is often worth more than a marginal capability gain from a preview checkpoint.
Preview ids shift. A prompt that produced a clean unified diff last week can sequence tools differently after a silent checkpoint update, and you will spend hours deciding whether your prompt or the model changed. Pinning "model": "google/gemini-2.5-pro" in atlas.json removes that variable. Because Atlas computes a unified diff for every file edit and surfaces it for approval before writing, and Atlas snapshots file changes as git patches so edits can be diffed and rolled back, a stable model plus reviewable diffs gives you a genuinely repeatable loop. Stability is a feature, and Gemini 2.5 Pro has it where the Gemini 3 previews do not.
How do you use the 1M window of Gemini 2.5 Pro well in Atlas?
Spend the 1,048,576 token window of Gemini 2.5 Pro on reading before writing. Atlas drafts a plan in a read-only plan agent and asks before switching to a build agent, so run the plan agent first on large refactors and let the cheap $1.25 per Mtok input carry the exploration cost.
The plan agent is where the Gemini 2.5 Pro economics pay off. In plan mode Atlas reads, searches, and reasons without touching the working tree, which is exactly the phase that consumes input tokens. At $1.25 per Mtok, a deep read of a large codebase is affordable, and the 65,536 token output cap on gemini-2.5-pro is more than enough to hold the resulting plan. Only after you approve the plan does Atlas switch to a build agent and start spending the more expensive $10 per Mtok output budget. Splitting the session this way converts the model's main strength, cheap long-context reading, directly into a lower bill.
When should you pick a different model than Gemini 2.5 Pro?
Pick a different model than Gemini 2.5 Pro when the task needs the strongest available reasoning. Gemini 2.5 Pro has been superseded on capability by the Gemini 3 line, so hard reasoning tasks favor the newer checkpoints even at $2 per Mtok input rather than $1.25.
Gemini 2.5 Pro is a 2025 flagship being used in 2026, and the honest framing is that it trades capability for price and stability. On a gnarly multi-file architectural change, the Gemini 3 line is the stronger pick. Atlas lets you switch the active model and provider on the fly with favorites and recents, so you do not have to commit: use /models mid session to A/B gemini-2.5-pro against gemini-3-pro-preview on the same task and compare the diffs Atlas renders. For cheap background work, neither Pro tier is right, and a Flash tier id belongs in the small_model slot instead.
Setup
- 01Export GOOGLE_GENERATIVE_AI_API_KEY or GEMINI_API_KEY with an AI Studio key.
- 02Run `atlas models google` and check that gemini-2.5-pro is listed as a stable id.
- 03Add "model": "google/gemini-2.5-pro" to atlas.json.
- 04Use the Atlas plan agent first on large refactors so the 1M window is spent on reading before the build agent writes.
- 05Switch mid session with /models if you want to A/B it against gemini-3-pro-preview.
Frequently asked questions
- how much does gemini 2.5 pro cost per million tokens
- Gemini 2.5 Pro costs $1.25 per Mtok input and $10 per Mtok output. The input price is 37 percent cheaper than Gemini 3 Pro's $2 for the same 1,048,576 token context window.
- is gemini 2.5 pro still worth using in 2026
- Gemini 2.5 Pro remains a strong default for long-context coding work because of its $1.25 per Mtok input price and its stable GA id. It is superseded on capability by the Gemini 3 line, so hard reasoning tasks favor the newer checkpoints.
- how do i configure gemini 2.5 pro in atlas
- Export GOOGLE_GENERATIVE_AI_API_KEY or GEMINI_API_KEY with an AI Studio key, run `atlas models google` to confirm gemini-2.5-pro is listed, then add "model": "google/gemini-2.5-pro" to atlas.json.
- what is the context window and output limit of gemini 2.5 pro
- Gemini 2.5 Pro has a 1,048,576 token context window and a 65,536 token output cap. It was the model that brought that window and output ceiling to the Gemini Pro tier.
- gemini 2.5 pro vs gemini 3 pro for coding agents
- Gemini 2.5 Pro is cheaper to read with at $1.25 per Mtok against Gemini 3 Pro's $2, and it is a stable GA id rather than a preview. Gemini 3 Pro is stronger on hard reasoning, so use /models in Atlas to A/B them on the same task.
- can i switch from gemini 2.5 pro to another model without restarting atlas
- Yes. Atlas lets you switch the active model and provider on the fly with favorites and recents, and /models moves the session between gemini-2.5-pro and another id like gemini-3-pro-preview.
- how do i keep gemini 2.5 pro costs down in atlas
- Run the Atlas plan agent first on large refactors so the 1,048,576 token window is spent on $1.25 per Mtok reading before the build agent starts spending the $10 per Mtok output budget.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas vs Sourcery: Terminal AI Coding Agents in 2026
Compare Atlas and Sourcery for 2026. Atlas offers terminal-native AI coding with permission-gated tools and local embeddings. Sourcery provides Python-first LLM review and rules-based refactoring, starting at
Atlas vs Devin: AI Coding Agents Compared for 2026
Atlas and Devin offer distinct AI coding experiences in 2026. Atlas provides a terminal-native TUI with local control, while Devin is a cloud-managed engineer with SWE-1.7.
Atlas vs Poolside: Terminal AI Coding Agents in 2026
Compare Atlas and Poolside, two terminal AI coding agents in 2026. Atlas offers a free core and robust safety features, while Poolside provides open-weight Laguna models and on-prem deployment options.
Atlas vs JetBrains Junie in 2026: Terminal Agent vs IDE Debugger-Driving Agent
Atlas vs JetBrains Junie in 2026: Junie drives the IDE debugger and reports 61.6% resolved on SWE-Rebench, bundled from $10/mo. Atlas is a free, open source terminal agent with permission gates.
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 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 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.
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.