# Atlas with Qwen2.5-Coder 1.5B (Ollama) in 2026 > Qwen2.5-Coder 1.5B (Ollama) offers a substantial 32K token context window at no cost, making it a compelling free, self-hosted option for Atlas users. Qwen2.5-Coder 1.5B (Ollama) is an excellent choice for developers in 2026 seeking a free, self-hosted model to handle Atlas's title generation and summarization tasks. Its 986MB download size allows it to run efficiently on integrated GPUs or CPUs, making it a cost-effective local option for specific agent roles within Atlas. ## Key takeaways - Qwen2.5-Coder 1.5B (Ollama) is Free (self-hosted) for Atlas users. - It provides a substantial 32K tokens (32,768) context window. - The model's 986MB download fits in roughly 3GB of VRAM or unified memory. - Best suited for Atlas's title generation and summarization tasks, not as the primary build agent. - Instruction following degrades under long tool-call chains; use 'ask' permissions in Atlas. - Supports local Ollama embeddings, keeping your code off third-party servers. ## What is Qwen2.5-Coder 1.5B (Ollama) best for in Atlas? Qwen2.5-Coder 1.5B (Ollama) excels as the designated small_model slot within Atlas, primarily handling title generation and summarization tasks in 2026. With a compact 986MB download, it efficiently loads on integrated GPUs or plain CPUs, freeing up resources for heavier models to manage complex editing workflows. This model is specifically designed to be the smallest Qwen2.5-Coder size that remains effective within an agent loop. Its efficiency means it can run locally, allowing Atlas to build its code index with local Ollama embeddings, thereby keeping your sensitive code off third-party servers. While a larger model performs the actual code editing, Qwen2.5-Coder 1.5B (Ollama) provides quick, local context and summaries, enhancing the overall Atlas experience without incurring external costs or data transfer overheads. ## What are the costs and context window for Qwen2.5-Coder 1.5B (Ollama)? Qwen2.5-Coder 1.5B (Ollama) is entirely Free (self-hosted) for developers in 2026, offering a generous 32K tokens (32,768) context window. This consistent context window across all Qwen2.5-Coder sizes ensures that prompts written for larger models can run unchanged on this 1.5B variant. The primary advantage of Qwen2.5-Coder 1.5B (Ollama) is its zero cost, as it operates entirely on your local hardware. This self-hosted nature means no API fees or subscription charges. Despite its small size, it maintains the same 32K token context window as its larger Qwen2.5-Coder counterparts, providing ample space for code context. The model's 986MB download on the default Q4_K_M tag means it fits comfortably within approximately 3GB of VRAM or unified memory, leaving room for the KV cache and making it accessible even on modest hardware configurations. ## When should I use a different model than Qwen2.5-Coder 1.5B (Ollama)? While Qwen2.5-Coder 1.5B (Ollama) is excellent for specific tasks, its 1.5B parameter count means it will struggle with complex multi-file refactors and long tool-call chains in 2026. Developers should treat it as the cheap slot for summarization, not the primary build agent for extensive code modifications. The inherent tradeoff with a model of this size is its limited capacity for deep reasoning and complex instruction following. When Atlas needs to perform multi-file refactors or engage in extended sequences of tool calls, Qwen2.5-Coder 1.5B (Ollama) may lose context or degrade in its instruction adherence. For these scenarios, it is crucial to pair Atlas with a larger, more capable model as the main 'model' setting. Furthermore, due to the potential for instruction following degradation under long tool-call chains, it is recommended to keep Atlas permissions on 'ask' rather than 'allow' when using Qwen2.5-Coder 1.5B (Ollama), ensuring human oversight for critical actions. ## How does Qwen2.5-Coder 1.5B (Ollama) support local Atlas workflows? Qwen2.5-Coder 1.5B (Ollama) is a cornerstone for local Atlas workflows in 2026, enabling the agent to build its code index with local Ollama embeddings. This ensures that all code and vector data remain on your machine, enhancing privacy and control, with a 986MB download footprint. Atlas is designed to be terminal-native and supports local operations extensively. By integrating Qwen2.5-Coder 1.5B (Ollama) via Ollama, Atlas can leverage local embeddings for its code indexing, a process that uses AST declarations via tree-sitter. This capability means your proprietary code never leaves your local environment, addressing significant privacy concerns. Ollama also ships the full quant matrix for this model size, from q2_K through fp16, along with base and instruct variants. This flexibility allows you to fine-tune the balance between model quality and memory footprint without needing to switch to a different model family, optimizing performance for your specific hardware and use case. ## Setup 1. Install Ollama on your system. 2. Run the command: ollama pull qwen2.5-coder:1.5b (986MB) to download the model. 3. Confirm Atlas sees your hardware by running: atlas device (it reports whether Ollama is installed and whether a GPU is present). 4. In your atlas.json configuration file, add an ollama provider whose npm field is @ai-sdk/openai-compatible and whose options.baseURL is http://localhost:11434/v1. 5. Register the model key qwen2.5-coder:1.5b within the ollama provider, setting limit.context to 32768 and limit.output to 4096. 6. Point your Atlas configuration's small_model field at ollama/qwen2.5-coder:1.5b and leave the main model field on something larger, ensuring the 1.5B only handles titles and summaries. 7. Run atlas models ollama to verify model resolution, then index your repository with Ollama embeddings so both the code and the vectors stay on your machine. ## FAQ ### What is the cost of Qwen2.5-Coder 1.5B (Ollama) for Atlas? Qwen2.5-Coder 1.5B (Ollama) is Free (self-hosted) for Atlas users, requiring only local hardware resources and incurring no API fees. ### What is the context window size for Qwen2.5-Coder 1.5B (Ollama)? Qwen2.5-Coder 1.5B (Ollama) provides a 32K tokens (32,768) context window, which is consistent across all Qwen2.5-Coder sizes, allowing for extensive code context. ### Can Qwen2.5-Coder 1.5B (Ollama) handle complex coding tasks in Atlas? No, as a 1.5B model, Qwen2.5-Coder 1.5B (Ollama) will lose the thread on multi-file refactors; it is intended as the cheap slot for summarization and title generation, not the primary build agent for complex edits. ### How much memory does Qwen2.5-Coder 1.5B (Ollama) require? The default Q4_K_M tag for Qwen2.5-Coder 1.5B (Ollama) is a 986MB download, fitting in roughly 3GB of VRAM or unified memory and leaving room for the KV cache. ### Does Atlas keep my code local with Qwen2.5-Coder 1.5B (Ollama)? Yes, Atlas can build its code index with local Ollama embeddings when using Qwen2.5-Coder 1.5B (Ollama), ensuring your code and vector data remain entirely on your machine, off third-party servers. ### What are the instruction following limitations of Qwen2.5-Coder 1.5B (Ollama)? Instruction following degrades under long tool-call chains with Qwen2.5-Coder 1.5B (Ollama), so it is recommended to keep Atlas permissions on 'ask' rather than 'allow' to maintain control over agent actions. ### Can I switch models in Atlas on the fly? Yes, Atlas lets you switch the active model and provider on the fly with favorites and recents, enabling you to use Qwen2.5-Coder 1.5B (Ollama) for specific lightweight tasks and a larger model for more demanding operations. --- Canonical HTML: https://seashell.sh/resources/models/qwen2-5-coder-1-5b-local Source of truth: aeo_pages row `/resources/models/qwen2-5-coder-1-5b-local` (segment: Models) (this file is generated from it, never hand-edited). Licence: SeaShell is proprietary with a free core. It is not open source and there is no public source repository.