Atlas integrates GLM-4.5, Z.ai's 2025 flagship model, offering a robust 128K token context window and a high 98,304 token output cap for extensive code generation and rewrites. Priced at $0.60 per Mtok input and $2.20 per Mtok output, it provides a cost-effective entry point for complex reasoning tasks, especially given its MIT license for flexible deployment.
What is GLM-4.5 best for in Atlas?
GLM-4.5, Z.ai's 2025 flagship model, excels within Atlas for tasks requiring substantial output and flexible deployment, thanks to its 98,304 max output tokens. This capacity supports whole-file rewrites in a single pass, making it ideal for refactoring large codebases or generating extensive documentation.
When driving Atlas, GLM-4.5 is particularly effective for scenarios demanding significant code generation or modification. Its unusually high 98,304 max output tokens, a standout feature for a 2025-era model, allows Atlas to perform comprehensive file rewrites without needing multiple turns. This is crucial for tasks like large-scale refactoring, boilerplate generation, or detailed documentation updates. Furthermore, GLM-4.5's MIT license provides unparalleled flexibility; developers can commercially use, redistribute, and privately fine-tune the model without vendor lock-in. This makes it an excellent choice for organizations that prioritize data privacy or require custom adaptations of their AI coding agent. Atlas can leverage GLM-4.5's reasoning capabilities, enabled at $0.60 per Mtok input, to draft detailed plans in its read-only plan agent before executing changes, ensuring thoughtful and permission-gated operations.
What are the cost and context tradeoffs of GLM-4.5?
GLM-4.5 offers a 128K token (131,072) context window at a competitive price of $0.60 per Mtok input and $2.20 per Mtok output. However, in 2026, this context size is the smallest in the live GLM lineup, with newer models like GLM-4.6 and GLM-4.7 providing 204,800 tokens.
While GLM-4.5 provides a substantial 128K token (131,072) context window, it is important to note that this is the smallest in the current GLM family as of 2026. Both GLM-4.6 and GLM-4.7 have expanded their context windows to 204,800 tokens, offering more capacity for extremely large codebases or complex, multi-file tasks. The pricing for GLM-4.5 is set at $0.60 per Mtok for input and $2.20 per Mtok for output. This input price is notable as it matches the much later GLM-4.7, establishing a price floor for the family's reasoning capabilities. The primary tradeoff is that GLM-4.6 offers an identical pricing structure ($0.60/$2.20) but with a larger context window and improved coding scores. Therefore, GLM-4.5 primarily serves as a pinned, reproducible checkpoint for those who require a specific, stable version of the model, rather than the absolute latest in performance or context capacity. Atlas users can switch models on the fly, allowing for easy comparison and selection based on specific task requirements and budget.
When should I choose a different model over GLM-4.5?
In 2026, you should consider alternatives to GLM-4.5 if your Atlas workflows demand the largest possible context window or the absolute best coding performance, as GLM-4.6 offers a 204,800 token window and better coding scores at the same $0.60/$2.20 price point.
While GLM-4.5 is a capable model, especially with its high output token limit and MIT license, there are specific scenarios where other models, particularly within the GLM family, might be a better fit for Atlas. If your development tasks frequently involve extremely large codebases or require processing an extensive amount of contextual information beyond 131,072 tokens, GLM-4.6 or GLM-4.7, with their 204,800 token context windows, would provide superior capacity. Furthermore, if your priority is the highest possible coding performance scores, GLM-4.6 is documented to offer better results than GLM-4.5, while maintaining the identical pricing of $0.60 per Mtok input and $2.20 per Mtok output. GLM-4.5's main distinction in 2026 is its role as a stable, reproducible checkpoint. If reproducibility across specific model versions is not a critical requirement, and you seek either greater context or enhanced coding performance, exploring GLM-4.6 or GLM-4.7 through Atlas's on-the-fly model switching feature would be advisable.
Can I run GLM-4.5 locally with Atlas?
Yes, you can run GLM-4.5 locally with Atlas by pulling the zai-org/GLM-4.5 weights from HuggingFace, but be aware that its 355B total parameters make it a multi-GPU deployment, not suitable for a single laptop.
Atlas supports running GLM-4.5 locally, leveraging its open-weights nature. To do this, you would pull the zai-org/GLM-4.5 model from HuggingFace. This allows you to keep your code entirely off third-party servers, enhancing privacy and control, especially when Atlas builds its code index with local Ollama embeddings. However, it is crucial to understand the hardware requirements: with 355B total parameters and 32B active parameters, GLM-4.5 is designed for multi-GPU deployments. It is not a model that can be run effectively on a typical laptop. For developers and teams with the necessary infrastructure, local deployment offers the full benefits of the MIT license, including private fine-tuning and complete control over the model's environment, integrating direct with Atlas's permission-gated tool calls and local git operations.
Setup
- 01Obtain an API key from Z.ai for access to GLM-4.5.
- 02Set your Z.ai API key by exporting it as ZHIPU_API_KEY in your environment, or run `atlas login` and select Z.ai from the provider list.
- 03Verify that `glm-4.5` is resolvable from the registry by executing the command `atlas models zai`.
- 04Configure Atlas to use GLM-4.5 by adding `"model": "zai/glm-4.5"` to your `atlas.json` configuration file.
- 05For local deployment, pull the `zai-org/GLM-4.5` weights from HuggingFace, noting that this 355B total parameter model requires a multi-GPU setup.
Frequently asked questions
- What is the context window size for GLM-4.5 in Atlas?
- GLM-4.5 provides a 128K token (131,072) context window when used with Atlas, allowing for substantial code and context processing.
- How much does it cost to use GLM-4.5 with Atlas?
- Using GLM-4.5 with Atlas costs $0.60 per Mtok for input tokens and $2.20 per Mtok for output tokens.
- Can GLM-4.5 handle large code rewrites in Atlas?
- Yes, GLM-4.5 is well-suited for large code rewrites in Atlas due to its unusually high 98,304 max output tokens, enabling whole-file modifications in one pass.
- Is GLM-4.5 an open-weights model?
- Yes, GLM-4.5 is an open-weights model from Z.ai, released under the MIT license, which permits commercial use, redistribution, and private fine-tuning.
- Why might I choose GLM-4.6 over GLM-4.5 for Atlas?
- In 2026, GLM-4.6 offers a larger 204,800 token context window and better coding scores at the identical $0.60/$2.20 price point, making it a stronger choice if maximum context or performance is your priority over a pinned checkpoint.
- What are the hardware requirements for running GLM-4.5 locally?
- Running GLM-4.5 locally requires a multi-GPU deployment due to its 355B total parameters; it is not designed for single-laptop use.
- How does Atlas leverage GLM-4.5's capabilities for code search?
- Atlas uses its Axis, the hybrid semantic and keyword code search fused by reciprocal rank fusion, to search code, which can then be fed into GLM-4.5's context for reasoning and generation.
Try SeaShell in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install SeaShellRelated guides
Atlas vs Sourcery: Terminal AI Coding Agents in 2026
Comparing Atlas, the terminal-native AI coding agent, with Sourcery, the Python-first AI reviewer, in 2026. Evaluate features, pricing, and workflow for developers.
Atlas for WebAssembly: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for WebAssembly. Audit wasm-bindgen exports, batch calls across the JS boundary, and run wasm-pack test --node in 2026.
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.
Atlas vs Mistral Vibe for Code: Terminal AI Coding Agents in 2026
Compare Atlas and Mistral Vibe for Code in 2026. Atlas offers terminal-native TUI, permission-gated tools, and local embeddings. Mistral Vibe provides a four-model stack and EU data sovereignty.
Atlas for Java in 2026
Adopt Atlas, the terminal-native AI coding agent, for Java development in 2026. Enhance your workflow with intelligent code search, refactoring, and robust safety features for Maven and Gradle projects.
Atlas for Groovy: A Terminal-Native AI Coding Agent for Gradle, Spock, and Jenkins in 2026
Atlas is a terminal-native AI coding agent for Groovy in 2026. It reads build.gradle closures and Jenkinsfiles, writes Spock specs, runs ./gradlew test, and applies Spotless.
Atlas vs OpenAI Codex CLI: A Developer's Guide for 2026
Compare Atlas and OpenAI Codex CLI for 2026. Atlas offers terminal-native TUI, BYO models, and granular change review, while OpenAI Codex CLI provides sandboxed execution and OpenAI model integration.
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.