MiniMax M2.5 and GLM-5 are now in Kiro

We’ve been expanding Kiro’s native support for open weight models, most recently with MiniMax M2.5 and now GLM-5, both directly available in the Kiro IDE and CLI. Kiro already supports a range of models across the cost-context-speed spectrum. These two additions extend that range further, giving developers and teams more room to choose based on the work in front of them.

The models

Here’s a closer look at what each model brings and where we think they’re good at.

MiniMax M2.5 (0.25x credit multiplier) — A sparse MoE model that activates 10B parameters per query. At just a quarter credit cost, it scores 80.2% on SWE-Bench Verified, the first open weight model to surpass Claude Sonnet and behind only Claude Opus 4.6 (80.8%), making it one of the most cost-efficient models in Kiro. It completes complex agentic tasks 37% faster than MiniMax M2.1. Before writing code, the model decomposes features and maps structure, which makes it excel for multi-step implementation work and longer agentic sessions. It also delivers strong multilingual support across 10+ languages (including Go, C, C++, TypeScript, Rust, Kotlin, Python, Java, JavaScript, and more) and handles full-stack projects spanning Web, Android, iOS, and Windows. If you want a fast, cost-efficient model for sustained coding sessions and iterative implementation passes, MiniMax M2.5 is a strong option.

GLM-5 (0.5x credit multiplier) — This is a large MoE model featuring a 200K context window. GLM-5 is optimized for long-horizon agentic workflows. It excels at processing repository-scale context and maintaining coherence during multi-step tool use across large codebases. Think cross-file migrations, full-stack feature development, or legacy refactoring where you need the model to hold the full picture. If you’re working through complex architectural changes that benefit from deep context, GLM-5 is worth trying.

Try them out in the IDE and CLI

You can access these models now with experimental support in the IDE model selector and through the Kiro CLI. MiniMax M2.5 is available in the AWS US-East-1 (N. Virginia) and AWS EU-Central-1 (Frankfurt) regions; inference for GLM-5 runs in the AWS US-East-1 (N. Virginia) region.

We’ve also expanded access to open weight models already in Kiro: MiniMax M2.1, Qwen3 Coder Next, and Deepseek V3.2 are now available to all users, including those authenticating through IAM Identity Center (IdC). To bring inference closer to your workloads, MiniMax M2.1 and Qwen3 Coder Next are also available in AWS EU-Central-1 (Frankfurt) region alongside AWS US-East-1 (N. Virginia).

Pick a model and start working, no configuration, no routing, no extra setup. Switch between them, set a default for specific project types, or let Auto handle it — whatever fits your workflow. For enterprise teams, administrators can use model governance to align available models with their compliance and data residency requirements. As always, experiment and let us know how these are working for you. We’re paying close attention to which models resonate and what gaps remain. If there’s a model you’d like to see supported next, let us know.