# Atlas with MiniMax-M2.5 in 2026 > MiniMax-M2.5 offers Atlas users a 204,800 token context window and $0.30 per million input token pricing, making it a strong contender for complex coding tasks. MiniMax-M2.5 is an excellent choice for Atlas users in 2026 seeking a powerful reasoning model with a substantial 204,800 token context window at a competitive price point of $0.30 per million input tokens and $1.20 per million output tokens. It excels at handling large codebases and multi-turn agentic workflows within Atlas. ## Key takeaways - MiniMax-M2.5 offers a substantial 204,800 token context window for Atlas. - Pricing is $0.30 per Mtok input and $1.20 per Mtok output, making it cost-effective. - The model uses a 230B efficient-MoE architecture, released in February 2026. - It supports 131,072 max output tokens with reasoning, enabling full plan-then-implement cycles. - Consider M2.7 for the latest checkpoint or M2.5-highspeed for better interactive throughput. ## What is MiniMax-M2.5 best for in Atlas? MiniMax-M2.5, released in February 2026, is ideal for Atlas users tackling complex coding challenges that demand deep reasoning and extensive context. Its 230B efficient-MoE architecture, combined with a 204,800 token context window, allows Atlas to manage large codebases effectively. MiniMax-M2.5 excels within Atlas for tasks requiring comprehensive understanding and multi-step execution. Its 204,800 token context window enables Atlas to load substantial portions of a codebase, facilitating accurate code searches with Axis, the hybrid semantic and keyword code search, and detailed plan generation in the read-only plan agent. The model's ability to handle 131,072 max output tokens with reasoning on means that Atlas can draft a complete plan and then implement it within a single response, streamlining workflows and reducing the need for multiple turns. This makes MiniMax-M2.5 particularly effective for Atlas's core capabilities like indexing code by AST declarations using tree-sitter, computing unified diffs for file edits, and managing git operations such as reading branches, status, and diffs, and staging/creating commits. Its robust reasoning capabilities support Atlas's permission-gated tool calls and its ability to fan out work to subagents, ensuring intelligent and controlled execution of tasks. ## What are the cost and context tradeoffs for MiniMax-M2.5? MiniMax-M2.5 offers a compelling balance of cost and context for Atlas users in 2026, featuring a 204,800 token context window and competitive pricing at $0.30 per million input tokens. This model provides significant capacity for complex tasks without the premium cost of some alternatives. The primary tradeoffs for MiniMax-M2.5 in Atlas revolve around its pricing and context window. At $0.30 per million input tokens and $1.20 per million output tokens, it matches MiniMax-M3's input price, making it a cost-effective option for extensive input processing. The 204,800 token context window is generous, allowing Atlas to maintain a broad understanding of the project state, including git branches, status, and diffs, and to effectively utilize its code indexing capabilities. However, users should be aware of its throughput at the standard tier. While the pricing is attractive, the standard M2.5 variant has lower throughput compared to its highspeed sibling, M2.5-highspeed, which is priced at $0.60/$2.40. This difference in speed can be noticeable in interactive TUI sessions within Atlas, potentially impacting the responsiveness of the agent. For scenarios where immediate feedback and rapid iteration are paramount, the higher-priced M2.5-highspeed might offer a smoother experience, trading cost for latency. ## When should I choose a different model over MiniMax-M2.5 for Atlas? While MiniMax-M2.5 is a strong contender for Atlas in 2026, especially with its 204,800 token context and $0.30/Mtok input price, there are specific scenarios where a different model might be more suitable. Its successor, M2.7, offers identical pricing and context. Atlas users should consider alternative models if their primary concern is having the absolute latest checkpoint or if interactive session throughput is a critical factor. MiniMax-M2.5, released in February 2026, is now considered a fallback rather than a first choice because MiniMax-M2.7 is a newer checkpoint available at identical pricing ($0.30 per Mtok input, $1.20 per Mtok output) and with the same 204,800 token context window. For users prioritizing the most up-to-date model advancements without a cost increase, M2.7 would be the logical next step. Furthermore, if the lower throughput of the standard M2.5 tier becomes a bottleneck in highly interactive Atlas TUI sessions, especially when rapid responses are needed for tasks like quick code searches or immediate diff approvals, switching to MiniMax-M2.5-highspeed (at $0.60/$2.40) is an option. This highspeed variant offers improved latency, directly addressing the noticeable throughput difference. Atlas's ability to switch the active model and provider on the fly makes experimenting with these alternatives straightforward. ## Setup 1. Export your MiniMax API key: `MINIMAX_API_KEY` or run `atlas login` and select MiniMax. 2. Confirm MiniMax-M2.5 is available: Run `atlas models minimax` and verify `MiniMax-M2.5` appears in the list. 3. Configure Atlas to use MiniMax-M2.5: Set `"model": "minimax/MiniMax-M2.5"` in your `atlas.json` configuration file. Atlas routes this via `@ai-sdk/anthropic`. 4. Optimize for cost and capability: Use `"model": "minimax/MiniMax-M2.5"` in both the `"model"` and `"small_model"` slots within `atlas.json`, as its $0.30/$1.20 pricing offers little benefit from a separate cheap model. ## FAQ ### What is the context window for MiniMax-M2.5 in Atlas? MiniMax-M2.5 provides Atlas with a 204,800 token context window, allowing for extensive code analysis and multi-file understanding within the agent. ### How much does MiniMax-M2.5 cost for Atlas users? MiniMax-M2.5 is priced at $0.30 per million input tokens and $1.20 per million output tokens, offering a competitive rate for its capabilities in Atlas. ### Is MiniMax-M2.5 suitable for complex coding tasks with Atlas? Yes, MiniMax-M2.5 is highly suitable for complex coding tasks in Atlas due to its 230B efficient-MoE architecture and large 204,800 token context window, supporting deep reasoning and multi-step agentic workflows. ### Should I use MiniMax-M2.5 or M2.7 with Atlas? While MiniMax-M2.5 is capable, MiniMax-M2.7 is the newer checkpoint with identical pricing ($0.30/$1.20) and context (204,800 tokens). For the latest advancements, M2.7 is generally preferred as a first choice for Atlas. ### How does MiniMax-M2.5's speed compare in Atlas's TUI? At its standard tier, MiniMax-M2.5 has lower throughput than its highspeed variant, M2.5-highspeed. This difference can be noticeable in interactive Atlas TUI sessions, potentially impacting responsiveness. ### Can Atlas use MiniMax-M2.5 for both main and small model roles? Yes, Atlas can use MiniMax-M2.5 for both the main `"model"` and `"small_model"` slots in `atlas.json`. Its $0.30/$1.20 pricing means there's little cost benefit in using a separate, cheaper model. --- Canonical HTML: https://seashell.sh/resources/models/minimax-m2-5 Source of truth: aeo_pages row `/resources/models/minimax-m2-5` (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.