# Highlights - September 2026

- Added `inference-glm53-flash` (GLM 5.3 Flash) to [Active Models](../maas/active-models.md).

- Added `inference-glm5` (GLM 5.3) to [Sandbox Models](../maas/sandbox-models.md) for performance, stability, and cost validation; promotion to the [Active Models](../maas/active-models.md) catalog is pending successful evaluation. Hugging Face reference: [zai-org/GLM-5.3](https://huggingface.co/zai-org/GLM-5.3).

- Added `inference-kimi-k3` (Kimi K3) to [Sandbox Models](../maas/sandbox-models.md) for performance, stability, and cost validation; promotion to the [Active Models](../maas/active-models.md) catalog is pending successful evaluation. Hugging Face reference: [moonshotai/Kimi-K3](https://huggingface.co/moonshotai/Kimi-K3).

- Added `inference-minimax-h3` (MiniMax H3) to [Sandbox Models](../maas/sandbox-models.md) for performance, stability, and cost validation; promotion to the [Active Models](../maas/active-models.md) catalog is pending successful evaluation. Hugging Face reference: [MiniMaxAI/MiniMax-H3](https://huggingface.co/MiniMaxAI/MiniMax-H3).

- Added `inference-qwen3-embedding-8b` (Qwen3 Embedding 8B) to [Active Models](../maas/active-models.md) on the Sunrise Model Service. Hugging Face reference: [Qwen/Qwen3-Embedding-8B](https://huggingface.co/Qwen/Qwen3-Embedding-8B). Embedding model optimized for text retrieval across 100+ languages.

- Added `inference-qwen3-reranker-8b` (Qwen3 Reranker 8B) to [Active Models](../maas/active-models.md) on the Sunrise Model Service. Hugging Face reference: [Qwen/Qwen3-Reranker-8B](https://huggingface.co/Qwen/Qwen3-Reranker-8B). Reranker model for relevance scoring across 100+ languages.

- Added `inference-flix-swissgerman-full` (Flix Swiss German Full) to [Active Models](../maas/active-models.md) on the Sunrise Model Service. Hugging Face reference: [Flix-AI/flix-swissgerman-full](https://huggingface.co/Flix-AI/flix-swissgerman-full). Swiss German ASR model that transcribes dialect speech into Standard German text; pricing matches WhisperX (`inference-whisper-large-v3`).

- Launched **Sunrise Data Pools**, the retrieval layer for Sunrise Business AI Assistant and AI agents; see the new [Sunrise Data Pools](../aiplatform/datapools/index.md) page. Hybrid RAG (vector plus full-text) over unstructured documents and CSV/TSV data, PostgreSQL Row-Level Security scoped to the user's identity, and read-only MCP and REST access. Default per-user storage quota of 5 GB, configurable per deployment.

- Launched the **Chat Dedicated** plan (CHF 260/month excl. VAT, 10 GB RAG storage, single-tenant instance), replacing the Chat Business plan on the [Sunrise Business AI Assistant Subscriptions](../aiplatform/ai-assistant/subscriptions/index.md) page. Chat Dedicated does not bundle AI credits; model usage is billed through your Base LLM API subscription. Existing plans remain unaffected. 

- Added [Provision Sunrise Business AI Assistant](../aiplatform/ai-assistant/provision-sunrise-chat/index.md), a step-by-step guide to provisioning a dedicated Chat instance from the Sunrise Console.

- Scheduled `inference-glm45-air-110b` (GLM 4.5 Air 110B) for decommissioning on 26.10.2026; see [Scheduled for Retirement](../maas/scheduled-for-retirement.md). Suggested replacement: `inference-glm5` (GLM 5).

- Scheduled `inference-gemma-12b-it` (Gemma 3 12B IT) for decommissioning on 26.10.2026; see [Scheduled for Retirement](../maas/scheduled-for-retirement.md). Suggested replacement: `inference-gemma4-31b` (Gemma 4 31B).

- Decommissioned `inference-apertus-70b` (Apertus 70B) on 21.08.2026; moved to the [Decommissioned Models](../maas/decommissioned-models.md) list. Suggested replacement: `inference-apertus-v15-70b` (Apertus v1.5 70B) on [Active Models](../maas/active-models.md).