Amazon Nova 2 Lite is the cheap, AWS-resident model for Atlas. It gives you a 128K token context at $0.33 / $2.75 per Mtok (input / output) on Bedrock, which makes it a strong fit for the background slot in Atlas, things like session titles and commit summaries, and a poor fit as the build-agent model on nontrivial code. Because it runs on the IAM role your team already has, no code leaves your AWS boundary.
What is Amazon Nova 2 Lite best at inside Atlas?
Amazon Nova 2 Lite is best used inside Atlas as the cheap background model, wired to the small_model slot for session titles and commit summaries. At $0.33 per Mtok input on Bedrock, Nova 2 Lite costs a third of Claude Haiku 4.5's $1 for that same background work.
Atlas runs a constant stream of low-stakes model calls behind the visible agent loop. Naming a session, summarizing a diff Atlas has already computed, and drafting a commit message when Atlas stages changes are all jobs where a 128K context and modest reasoning are perfectly sufficient. Amazon Nova 2 Lite is priced for exactly that tier. Wire it with "small_model": "amazon-bedrock/amazon.nova-2-lite-v1:0" in atlas.json and leave a stronger model in the main slot. The economics are the whole argument here: every token Atlas spends on housekeeping is a token you are not spending on the model actually writing code.
How much does Amazon Nova 2 Lite cost per million tokens?
Amazon Nova 2 Lite costs $0.33 / $2.75 per Mtok (input / output) on Amazon Bedrock in 2026. The $0.33 input rate is one third of Claude Haiku 4.5's $1, which is the entire reason to reach for Nova 2 Lite on high-volume, low-stakes Atlas traffic.
Pricing for Amazon Nova 2 Lite is worth reading carefully before you commit. The $0.33 input rate is genuinely cheap, but the $2.75 output rate is not trivially cheap, so Nova 2 Lite rewards workloads that read a lot and write a little. Atlas background tasks fit that shape: they consume a diff or a transcript and emit a sentence. Tasks that emit long code bodies do not fit it. Note also that the original Nova Lite is cheaper still at $0.06 / $0.24 with a larger 300K window, so Amazon Nova 2 Lite only pays for itself when you specifically want the generational quality gain.
Is a 128K context window enough for Atlas build-agent work?
A 128K token context window is not enough for Atlas build-agent work on nontrivial code, and Amazon Nova 2 Lite has exactly 128K. Nova 2 Lite pairs that window with modest reasoning, which is the combination that makes it unsuitable as the model driving edits.
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, plus a plan, plus a running edit history, fill a context window quickly. With 128K tokens, Amazon Nova 2 Lite will hold a focused single-file task and lose the thread on a multi-file refactor. Treat the 128K number as a hard planning constraint rather than a headline. If your Atlas work regularly spans several files and a long tool-call history, Nova 2 Lite is the wrong seat for it.
Why run Amazon Nova 2 Lite inside your AWS boundary?
Amazon Nova 2 Lite runs entirely within your AWS boundary on Bedrock, which satisfies data policies that forbid sending code to a third-party API. For regulated teams in 2026, that single property often outweighs the raw capability gap against a frontier model.
Procurement is a real constraint on model choice. Amazon Nova 2 Lite requires no separate vendor onboarding: it uses the IAM role your team already has, authenticated through AWS_PROFILE or AWS_ACCESS_KEY_ID plus AWS_SECRET_ACCESS_KEY, in the region you set with AWS_REGION. Nothing new to sign, no new data processing agreement. Pair that with Atlas's own local-first posture, since Atlas can build its code index with local Ollama embeddings, keeping code off third-party servers, and you get a setup where prompts stay in AWS and vectors stay on the machine. Every Atlas tool call is still permission-gated against allow, ask, and deny rules before it runs.
When should you pick a different model than Amazon Nova 2 Lite?
Pick a different model than Amazon Nova 2 Lite whenever Atlas is doing the actual building. Nova 2 Lite's 128K context and modest reasoning make it unsuitable as an Atlas build-agent model on nontrivial code, so reserve it for the cheap slot and let a stronger model hold the main slot.
There are two clean exits from Amazon Nova 2 Lite. The first is upward: when Atlas is planning a refactor, computing unified diffs across many files, or fanning work out to parallel subagents, move the main slot to a model with more headroom and keep Nova 2 Lite only in small_model. The second is sideways: if your only goal is the lowest possible cost per token and you do not need the generational quality gain, the original Nova Lite at $0.06 / $0.24 with a 300K window is cheaper and roomier. Atlas lets you switch the active model and provider on the fly with favorites and recents, so testing both is a keystroke.
Setup
- 01Set AWS_REGION and authenticate via AWS_PROFILE, or via AWS_ACCESS_KEY_ID plus AWS_SECRET_ACCESS_KEY.
- 02Request access to Nova 2 Lite in the Bedrock model catalog for your region.
- 03Confirm Atlas resolved the provider with `atlas models amazon-bedrock`.
- 04Wire Amazon Nova 2 Lite to the cheap slot: "small_model": "amazon-bedrock/amazon.nova-2-lite-v1:0" in atlas.json.
- 05Open /models in the Atlas TUI and confirm Amazon Nova 2 Lite appears, then keep a stronger model in the main slot for build work.
Frequently asked questions
- how much does amazon nova 2 lite cost per million tokens
- Amazon Nova 2 Lite is $0.33 / $2.75 per Mtok (input / output) on Amazon Bedrock. The input rate is one third of Claude Haiku 4.5's $1 per Mtok.
- what is the context window of amazon nova 2 lite
- Amazon Nova 2 Lite has a 128K token context window. That is enough for background Atlas jobs like session titles and commit summaries, but too small for build-agent work on nontrivial code.
- how do i use amazon nova 2 lite with atlas
- Set AWS_REGION, authenticate with AWS_PROFILE or AWS_ACCESS_KEY_ID plus AWS_SECRET_ACCESS_KEY, request access to Nova 2 Lite in the Bedrock model catalog, run `atlas models amazon-bedrock` to confirm, then set "small_model": "amazon-bedrock/amazon.nova-2-lite-v1:0" in atlas.json.
- is amazon nova 2 lite good enough to be the main coding model
- No. Amazon Nova 2 Lite's 128K context and modest reasoning make it unsuitable as an Atlas build-agent model on nontrivial code. Use it in the cheap slot and keep a stronger model in the main slot.
- nova 2 lite vs nova lite which is cheaper
- The original Nova Lite is cheaper, at $0.06 / $0.24 per Mtok with a larger 300K context window. Amazon Nova 2 Lite at $0.33 / $2.75 with 128K is only worth the premium if you want the generational quality gain.
- does amazon nova 2 lite keep my code inside aws
- Yes. Amazon Nova 2 Lite runs on Bedrock entirely within your AWS boundary, so it satisfies data policies that forbid sending code to a third-party API, and it uses the IAM role your team already has.
- what command lists bedrock models in atlas
- Run `atlas models amazon-bedrock`. It confirms which Bedrock models Atlas resolved, including Amazon Nova 2 Lite once you have requested access in the Bedrock model catalog for your region.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Atlas for Fortran: fpm.toml, Explicit Interfaces, and fprettify in 2026
Atlas is a terminal-native AI coding agent for Fortran in 2026. It reads modules, explicit interfaces, and intent declarations, runs fpm test behind a prompt, and runs fprettify.
Locate Where a Behavior Is Implemented with Atlas in 2026
How to locate where a behavior is implemented with Atlas in 2026: codebase_search for meaning, grep for exact text, and the lsp tool for the symbol graph.
Review a Pull Request with Atlas (2026 Workflow)
How to review a pull request with Atlas in 2026: bash produces the raw patch, read pulls whole files, the lsp tool's findReferences checks callers the diff never shows.
Atlas vs Roo Code: Terminal AI Coding Agents in 2026
Comparing Atlas and Roo Code in 2026. Atlas offers terminal-native TUI, permission-gated tools, and diff review. Roo Code, a VS Code extension, shut down May 15, 2026.
Atlas vs Base44: Terminal AI Coding Agents in 2026
Compare Atlas, the terminal-native AI coding agent, with Base44, the Wix-owned no-code app builder, for developers in 2026. Evaluate features, pricing, and workflow.
Atlas for Clojure: A Terminal-Native AI Coding Agent for deps.edn and Kaocha in 2026
Atlas is a terminal-native AI coding agent for Clojure in 2026. It reads deps.edn aliases and namespace requires, runs clojure -M:test with Kaocha, and applies cljfmt.
Atlas for React Native: Terminal-Native AI Coding Across the Native Boundary in 2026
Atlas is a terminal-native AI coding agent for React Native in 2026. Work across the New Architecture, native modules, and platform-specific files with diff-first review.
Atlas for Electron: Terminal-Native AI Coding for Main, Preload, and Renderer in 2026
Atlas is a terminal-native AI coding agent for Electron in 2026, where the main and renderer split, contextIsolation, and preload bridges are the security model.