What is the actual context window for Qwen2.5 72B (local via Ollama) in Atlas?

Qwen2.5 72B (local via Ollama) has a native context window of 128K tokens (131,072), but for local Atlas operations, it is commonly capped at 32,768 tokens due to KV cache memory constraints on consumer hardware.

How much does it cost to run Qwen2.5 72B (local via Ollama) with Atlas?

Running Qwen2.5 72B (local via Ollama) with Atlas is Free (self-hosted), meaning there are zero per-token costs. Your only investment is the initial hardware capable of running the model, such as a 64 GB Mac Studio.

Can Atlas use Qwen2.5 72B (local via Ollama) for local code embeddings?

Yes, Atlas can build its code index with local Ollama embeddings using OllamaEmbedder. This ensures that both your source code and its vector representations remain entirely on your machine when using Qwen2.5 72B (local via Ollama), maintaining full data privacy.

What hardware is needed to run Qwen2.5 72B (local via Ollama) with Atlas?

Qwen2.5 72B (local via Ollama) requires about 47 GB of VRAM at Q4_K_M quantization. This means a 64 GB Mac Studio or a system with a pair of 24 GB graphics cards can run the largest dense Qwen2.5 model effectively with Atlas.

How do I configure Atlas to use Qwen2.5 72B (local via Ollama)?

To configure Atlas, first pull the model with `ollama pull qwen2.5:72b`. Then, add an `ollama` custom provider block to `atlas.json` with `"baseURL": "http://localhost:11434/v1"` and register `"qwen2.5:72b"` with a context limit of 32768 tokens. Finally, set `"model": "ollama/qwen2.5:72b"`.

Does Qwen2.5 72B (local via Ollama) support Atlas's code search capabilities?

Yes, Qwen2.5 72B (local via Ollama) fully supports Atlas's code search capabilities, including Axis, the hybrid semantic and keyword retrieval fused by reciprocal rank fusion. This operates entirely locally, leveraging the model's understanding for effective code navigation.

Atlas with Qwen2.5 72B (local via Ollama) in 2026

Qwen2.5 72B (local via Ollama) is ideal for Atlas users prioritizing absolute data privacy and zero per-token cost in 2026. This self-hosted model allows Atlas to perform all coding agent tasks, including code indexing with OllamaEmbedder, entirely on your local hardware, ensuring nothing leaves your machine. It's completely Free (self-hosted) to run after your initial hardware investment, making it a compelling choice for secure, cost-controlled development environments.

What Qwen2.5 72B (local via Ollama) is Best At Inside Atlas

Qwen2.5 72B (local via Ollama) excels within Atlas for developers in 2026 who demand complete data sovereignty and zero operational costs. This 72B model, when paired with OllamaEmbedder, ensures that all code, vectors, and agent interactions remain entirely on your local machine, making it perfect for sensitive projects.

For Atlas users, Qwen2.5 72B (local via Ollama) provides a robust, fully air-gapped solution. Atlas can build its code index by AST declarations using tree-sitter, and with OllamaEmbedder, these embeddings stay local, keeping code off third-party servers. This setup is invaluable for projects with strict security or compliance requirements. Atlas's core capabilities, such as Axis, the hybrid semantic and keyword code search, and its permission-gated tool calls, operate entirely within your local environment. The model's ability to run on consumer hardware, specifically requiring about 47 GB at Q4_K_M, means a 64 GB Mac Studio or a pair of 24 GB graphics cards can host the largest dense Qwen2.5 model, enabling powerful local AI coding without external dependencies or per-token charges.

Cost and Context Window Tradeoffs with Qwen2.5 72B (local via Ollama)

While Qwen2.5 72B (local via Ollama) is entirely Free (self-hosted), offering zero per-token cost for Atlas agent loops, it presents specific tradeoffs for developers in 2026. The model's native 128K tokens (131,072) context window is commonly capped at 32,768 tokens locally due to KV cache limitations.

The primary tradeoff for Qwen2.5 72B (local via Ollama) within Atlas is the practical reduction of its impressive native context window. Although the model supports 128K tokens (131,072) natively, local deployments typically cap this at 32,768 tokens. This is because a full KV cache at the native length would not fit alongside the model weights on consumer hardware. Additionally, token throughput on such hardware will be limited to a few tokens per second. This can make long Atlas agentic runs, which often involve dozens of tool calls and extensive code analysis, significantly slower than with cloud-hosted alternatives. While the zero per-token cost is a major advantage, the time investment in waiting for responses must be factored into development workflows.

When to Pick a Different Model for Atlas

Developers in 2026 should consider alternative models for Atlas if their primary concern is raw speed or leveraging the full 128K token context window of Qwen2.5 72B. While Qwen2.5 72B (local via Ollama) is free, its token throughput on consumer hardware is limited to a few tokens per second, impacting the pace of complex agentic workflows.

If your Atlas workflow demands rapid iteration, or if you frequently require context windows exceeding the practical 32,768 tokens available locally, Qwen2.5 72B (local via Ollama) might not be the optimal choice. For scenarios where speed is paramount, or where the sheer volume of code and documentation necessitates a larger effective context window than what can be practically run on local hardware, a cloud-hosted model might offer a more efficient experience. Atlas lets you switch the active model and provider on the fly, making it easy to experiment with different options. If your local hardware cannot comfortably accommodate the approximately 47 GB model size, or if you prioritize immediate response times over absolute data privacy and zero per-token cost, exploring other models is advisable.

Setup

01Pull the Qwen2.5 72B model using Ollama: `ollama pull qwen2.5:72b` (this download is roughly 47 GB).
02Confirm the model loads and runs locally by executing: `ollama run qwen2.5:72b`.
03Add an `ollama` custom provider block to your `atlas.json` configuration file, specifying `"npm": "@ai-sdk/openai-compatible"` and `"baseURL": "http://localhost:11434/v1"`.
04Register the Qwen2.5 72B model within the `ollama` provider block, setting `"qwen2.5:72b"` with `"limit": { "context": 32768, "output": 8192 }`.
05Set `"model": "ollama/qwen2.5:72b"` in your Atlas configuration to make it the active model for your agent.
06Run `atlas device` to confirm Atlas successfully detects Ollama and your GPU hardware.
07Verify the Ollama provider is live and accessible to Atlas with the command: `atlas models ollama`.

Frequently asked questions

What is the actual context window for Qwen2.5 72B (local via Ollama) in Atlas?: Qwen2.5 72B (local via Ollama) has a native context window of 128K tokens (131,072), but for local Atlas operations, it is commonly capped at 32,768 tokens due to KV cache memory constraints on consumer hardware.
How much does it cost to run Qwen2.5 72B (local via Ollama) with Atlas?: Running Qwen2.5 72B (local via Ollama) with Atlas is Free (self-hosted), meaning there are zero per-token costs. Your only investment is the initial hardware capable of running the model, such as a 64 GB Mac Studio.
Can Atlas use Qwen2.5 72B (local via Ollama) for local code embeddings?: Yes, Atlas can build its code index with local Ollama embeddings using OllamaEmbedder. This ensures that both your source code and its vector representations remain entirely on your machine when using Qwen2.5 72B (local via Ollama), maintaining full data privacy.
What hardware is needed to run Qwen2.5 72B (local via Ollama) with Atlas?: Qwen2.5 72B (local via Ollama) requires about 47 GB of VRAM at Q4_K_M quantization. This means a 64 GB Mac Studio or a system with a pair of 24 GB graphics cards can run the largest dense Qwen2.5 model effectively with Atlas.
Is Qwen2.5 72B (local via Ollama) fast enough for Atlas's agentic workflows?: On consumer hardware, Qwen2.5 72B (local via Ollama) typically achieves a token throughput of only a few tokens per second. This can make long Atlas agentic runs, which involve dozens of tool calls, noticeably slower compared to cloud-hosted alternatives.
How do I configure Atlas to use Qwen2.5 72B (local via Ollama)?: To configure Atlas, first pull the model with `ollama pull qwen2.5:72b`. Then, add an `ollama` custom provider block to `atlas.json` with `"baseURL": "http://localhost:11434/v1"` and register `"qwen2.5:72b"` with a context limit of 32768 tokens. Finally, set `"model": "ollama/qwen2.5:72b"`.
Does Qwen2.5 72B (local via Ollama) support Atlas's code search capabilities?: Yes, Qwen2.5 72B (local via Ollama) fully supports Atlas's code search capabilities, including Axis, the hybrid semantic and keyword retrieval fused by reciprocal rank fusion. This operates entirely locally, leveraging the model's understanding for effective code navigation.

Try SeaShell in your terminal

The terminal-native AI coding agent. Free core, single binary.

Install SeaShell

Related guides

Self-Review Your Working Diff Before Committing with Atlas (2026 Workflow)

How to self-review your working diff before committing with Atlas in 2026: bash produces the diff, read checks each file, grep finds leftovers, session revert undoes bad edits.

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.

Atlas for React in 2026

Adopt Atlas, the terminal-native AI coding agent, for React development in 2026. Enhance your workflow with intelligent code search, refactoring, and testing for React components and hooks.

Atlas vs Tabby: Terminal AI Coding Agents in 2026

Comparing Atlas and Tabby in 2026: Atlas offers terminal-native AI with explicit planning and diffing, while Tabby provides self-hosted GPU-based completion and a cloud agent.

Atlas for Symfony in 2026

Atlas is a terminal-native AI coding agent for Symfony in 2026. It reads autowired services and Doctrine mappings, and shows migration SQL before anything runs.

Atlas for TypeScript in 2026

In 2026, TypeScript developers leverage Atlas, the terminal-native AI coding agent, to enhance productivity. Atlas understands your types, ensures code quality, and offers robust safety features.

Atlas vs Blackbox AI: Terminal AI Coding Agents in 2026

Comparing Atlas and Blackbox AI in 2026. Atlas offers terminal-native TUI, permission-gated tools, and local embeddings. Blackbox AI features /multi-agent dispatch and VS Code integration.

Atlas for Angular in 2026

Adopt Atlas, the terminal-native AI coding agent, for your Angular projects in 2026. Enhance development with intelligent code search, secure local embeddings, and granular control over AI actions.

Browse this resource hub