Qwen2.5-Coder 7B (local via Ollama) is the small, dependable local coder. At 7B with a 32K token window it runs on almost any modern laptop, in about 5GB at 4-bit quantization, and it is the model the reference Atlas local configuration ships with. It costs nothing to self-host. The tradeoff is the 32,768 token context, an order of magnitude below Qwen3-Coder's 256K, which large files and long sessions overflow quickly.
Can you run Atlas on a laptop with no discrete GPU?
Yes, with Qwen2.5-Coder 7B. Qwen2.5-Coder 7B runs comfortably in about 5GB at 4-bit quantization, so it works on a laptop with no discrete GPU. That 5GB footprint is the lowest bar of any model Atlas treats as a serious local default in 2026.
Most local coding models assume a GPU. Qwen2.5-Coder 7B does not, and that is its entire reason to exist in the Atlas lineup. Pull it with `ollama pull qwen2.5-coder:7b` and you have a working agent on a machine that cannot host anything larger. Atlas behavior is unchanged on this hardware: Atlas indexes code by AST declarations using tree-sitter, not blind line windows, and every Atlas tool call is permission-gated against allow, ask, and deny rules before it runs. The model is smaller, the harness is not.
Why is Qwen2.5-Coder 7B the reference local configuration for Atlas?
Qwen2.5-Coder 7B is the exact model that Atlas's reference local configuration pins for both `model` and `small_model`. Pinning one 7B model to both slots means a single 5GB download gets you a fully working local Atlas, with no second model to pull or configure.
Reference configurations are chosen for the widest hardware, not the best score. Setting both slots to Qwen2.5-Coder 7B, with "model": "ollama/qwen2.5-coder:7b" and "small_model": "ollama/qwen2.5-coder:7b" in atlas.json, means the background jobs and the build agent share one loaded model, which matters enormously on a machine with limited memory. Loading two separate models on a laptop would evict one from memory constantly. One model in both slots avoids that entirely.
Is a 32K context window enough for Atlas?
A 32,768 token context window is tight for Atlas. Qwen2.5-Coder 7B provides exactly 32K, an order of magnitude below Qwen3-Coder's 256K, so large files and long sessions overflow quickly and you will feel it within a few dozen turns.
Plan your Qwen2.5-Coder 7B sessions around the 32,768 token limit rather than fighting it. Keep tasks narrow: one file, one function, one bug. Register the honest limit in atlas.json with "limit": { "context": 32768, "output": 8192 } so Atlas trims deliberately instead of the model failing silently. Because Atlas snapshots file changes as git patches so edits can be diffed and rolled back, a session that goes off the rails from context overflow is cheap to undo, but it is cheaper still to keep the scope small in the first place.
What is the upgrade path from Qwen2.5-Coder 7B?
Qwen2.5-Coder is available at 0.5B, 1.5B, 3B, 7B, 14B, and 32B, giving a clean upgrade path as hardware improves. Moving from Qwen2.5-Coder 7B to the 14B or 32B variant is a one-line change in atlas.json, with no prompt rewriting required.
The six-size ladder is the practical strength of the Qwen2.5-Coder family. A developer on a 2020 laptop can run the 7B or drop to 3B, and the same developer on new hardware next year can pull the 32B without changing how they work with Atlas. Run `atlas models ollama` after any size change to confirm what Atlas resolved. When the hardware allows something better than the family, however, the honest recommendation is to leave it: qwen3-coder:30b beats Qwen2.5-Coder 7B decisively.
When should you pick a different model than Qwen2.5-Coder 7B?
Pick a different model than Qwen2.5-Coder 7B as soon as the hardware allows. Qwen2.5-Coder 7B is a previous-generation model, and qwen3-coder:30b beats it decisively wherever the hardware allows, with a 256K window against Qwen2.5-Coder 7B's 32,768 tokens.
Qwen2.5-Coder 7B earns its place on constrained machines and nowhere else. If you have a 24GB GPU or a 32GB Apple Silicon machine, Qwen3-Coder 30B is the better default by a wide margin, and the context difference alone, 256K against 32K, changes what Atlas can attempt. Atlas lets you switch the active model and provider on the fly with favorites and recents, so keep Qwen2.5-Coder 7B configured as the fallback for travel or a thin client, and run something larger when you are at your real workstation.
Setup
- 01Run `ollama pull qwen2.5-coder:7b` (about 5GB at 4-bit quantization).
- 02Add the provider block to atlas.json: "ollama": { "npm": "@ai-sdk/openai-compatible", "name": "Ollama (local)", "options": { "baseURL": "http://localhost:11434/v1" }, "models": { "qwen2.5-coder:7b": { "name": "Qwen2.5 Coder 7B (local)", "limit": { "context": 32768, "output": 8192 } } } }
- 03Set both slots: "model": "ollama/qwen2.5-coder:7b" and "small_model": "ollama/qwen2.5-coder:7b".
- 04Run `atlas models ollama` to confirm, then start the TUI.
- 05Keep tasks narrow, since the 32,768 token window overflows quickly on large files.
Frequently asked questions
- can i run atlas on a laptop without a gpu
- Yes, with Qwen2.5-Coder 7B. It runs comfortably in about 5GB at 4-bit quantization, so it works on a laptop with no discrete GPU. Pull it with `ollama pull qwen2.5-coder:7b`.
- what is the context window of qwen2.5-coder 7b
- Qwen2.5-Coder 7B has a 32,768 token context window, an order of magnitude below Qwen3-Coder's 256K. Large files and long Atlas sessions overflow it quickly, so keep tasks narrow.
- what is the default local model in atlas reference config
- Qwen2.5-Coder 7B is the exact model Atlas's reference local configuration pins for both `model` and `small_model`, so a single 5GB download gets you a fully working local Atlas.
- what sizes does qwen2.5-coder come in
- Qwen2.5-Coder is available at 0.5B, 1.5B, 3B, 7B, 14B, and 32B, giving a clean upgrade path as hardware improves. The 7B is the one that fits a laptop with no discrete GPU.
- qwen2.5-coder 7b vs qwen3-coder 30b
- Qwen2.5-Coder 7B is a previous-generation model, and qwen3-coder:30b beats it decisively wherever the hardware allows. The context gap is stark: 32,768 tokens against 256K.
- how do i configure qwen2.5-coder 7b in atlas.json
- Add the ollama provider with baseURL http://localhost:11434/v1, register "qwen2.5-coder:7b" with "limit": { "context": 32768, "output": 8192 }, then set both "model": "ollama/qwen2.5-coder:7b" and "small_model": "ollama/qwen2.5-coder:7b".
- how much does qwen2.5-coder 7b cost
- Qwen2.5-Coder 7B is free, self-hosted through Ollama. The only cost is the roughly 5GB download at 4-bit quantization and the laptop you already own.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
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 Axum in 2026
Atlas is a terminal-native AI coding agent for Axum in 2026. It decodes tower trait-bound errors, adds IntoResponse types, and runs cargo nextest run.
Atlas vs Amazon Q Developer: Terminal AI Coding Agents in 2026
Comparing Atlas and Amazon Q Developer in 2026. Atlas offers terminal-native AI with permission-gated tools and local code indexing. Amazon Q Developer excels in AWS services, providing Java upgrades and security
Atlas for Three.js: Terminal-Native AI Coding for Scenes, Materials, and Disposal in 2026
Atlas is a terminal-native AI coding agent for Three.js in 2026, where geometry, material, and texture disposal is the difference between a demo and a leak.
Atlas vs JetBrains AI Assistant: Terminal AI Coding Agents in 2026
Atlas and JetBrains AI Assistant comparison for 2026. Atlas offers a terminal-native TUI and BYO model keys, while JetBrains AI Assistant integrates into IDEs with credit-based cloud models.
Diagnose a Hanging or Long-Running Command with Atlas in 2026
How to diagnose a hanging command with Atlas in 2026: the bash tool races every command against a timeout and tells you whether it is slow or blocked on input.
Atlas vs Augment Code: Choosing a Terminal AI Coding Agent in 2026
In 2026, Atlas offers a free, terminal-native AI agent with local embeddings and robust safety. Augment Code's Cosmos platform provides shared memory for teams at $100/month.
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.