Microsoft Licenses Now in the Software Shop
As of now, there is an important update for IT purchasers. You can order items via the Microsoft Select Plus volume licensing agreement. This takes place conveniently in the asknet software shop. Specifically, these items are known as B-products.
Currently, the offer includes licenses without Software Assurance. Furthermore, you can purchase Software Assurance separately as maintenance. Do you want to order licenses that include Software Assurance instead? This is expected to be possible starting in late August. Therefore, we ask for a little patience.
Despite these updates, the existing framework conditions remain completely unchanged.
New Model Available for Employees and Students: GPT-5.5
Features:
- For employees and students, the model GPT-5.5 has been added. Data processing for GPT-5.5 takes place exclusively within the European Union.
- For employees, the following usage limits apply:
- GPT-5.5 (Reasoning high): 100 requests per 3 hours
- GPT-5.5 (Reasoning minimal): 100 requests per hour
- For students, the following usage limits apply:
- GPT-5.5 (Reasoning high): 100 requests per day
- GPT-5.5 (Reasoning minimal): 100 requests per 6 hours
- For GPT-5.4-mini, the usage limits have been adjusted as follows:
- Employees: 1,000 requests per hour
- Students: 250 requests per 6 hours
- The following models have been removed:
- GPT-5.2
- GPT-5.3
Release Notes Version 2.55.0
Improvements and bug fixes:
- Various bug fixes
GPU TDP reduced on CLAIX 2023 nodes
As of the 30.06.2026 at 11:30 the maximum TDP of the H100 GPUs on Claix 2023 nodes has been reduced from the previous value of 700W to 600W.
We implemented this lower TDP to reduce the thermal load on the central cooling system. This central cooling system was operating hotter and causing some specific GPU nodes to trigger thermal protection shutdowns that in turn kill any running jobs on said nodes.
This change is noteworthy for users that perform performance measurements and would like to compare their previous measurements to new ones under the reduced TDP.
For normal usage of the GPUs of CLAIX 2023, this change should not affect validity of results and should provide close to stock times to solution.
Users should not expect longer runtimes in most cases. For extremely intense computations, please recheck if your workflows require more runtime and adjust your batch scripts accordingly. We nevertheless, do not expect this change to affect most users of GPUs and is of no consequence to non-GPU users.
Jupyter
The NFDI4ING Jupyter Service team at the University of Stuttgart has successfully tested and documented the integration of an S3 resource from Coscine into the Jupyter Service. This makes it possible to interactively process Coscine data in reusable and documented workflows in the form of Jupyter Notebooks. We will soon add a note regarding this to the Coscine documentation. If you have an S3 resource in Coscine, you can already test the link to Jupyter Notebook.
First Production Release After Public Testing Phase [1.0.0]
First Major Release of the Production-Ready Product Following Completion of the Public Testing Phase
Features:
- Query Planner – RAG Retrieval Overhaul: A lightweight LLM call (gpt-5.4-mini) now runs before document retrieval to generate a cleaned, topic-focused search query, replacing the previous embedding-based topic-shift detection heuristic. Result: more accurate RAG retrieval and elimination of topic contamination (e.g., VPN and HPC being merged into a single incorrect query). Fully configurable at runtime through a new tab in the Env Manager, with a robust fallback ensuring that planner failures or timeouts never interrupt a chat session.
- Domain-Specific Search Language: HPC-/cluster-related queries are searched in English (matching the language of the knowledge base), while responses remain in the user’s preferred language.
- Explicit Language Switching: Instructions such as “answer in English” or “auf Deutsch bitte” now switch the response language reliably.
- More Accurate Incident Detection: Incident banners are displayed only when users explicitly ask about a disruption, rather than whenever a service is merely mentioned.
- Screenshot-Upload (Vision Support): Images can be attached via paste, file selection, or drag-and-drop. The client downsizes uploads to WebP, while the server validates and re-encodes image data using Pillow. Screenshots remain available throughout follow-up questions and are automatically removed after a configurable retention period.
- Dark Mode: New theme system with a dedicated dark stylesheet and theme toggle.
- Per-Message Test Cases and promptfoo Export: Administrators can mark individual question–answer turns as test cases, define the expected answer and correct documentation references, and export everything as promptfoo-compatible YAML files for automated quality evaluation.
Intern/Architektur
- The monolithic bot.py (775 lines) has been refactored into a structured app/bot/ package (pipeline, client, search, language, tokens, incidents, query_planner).
- routes.py has been modularized into subpackages under app/routes/.
- Integrated deployment webhook (seko_deploy_webhook), expanded test coverage, and CI/CD updates.
Bugfixes and Improvements:
- Source Attribution: Improved tracking of RAG source documents. Shared chats now display the referenced documentation under a dedicated Sources section.
- Feedback Review System Overhaul: Refactored review interface, introduced a flag queue for suspicious chats, and normalized the database schema using enums and lookup tables for feedback categories and AI models.
- Env Manager Redesign: Database-backed configuration management with bidirectional .env-synchronization, side-by-side configuration views, a dedicated write user, and export of merged .env files.
- Incident Processing: Improved prioritization by severity, time-based filtering, token-budget-aware truncation, and clickable incident links.
- Updated GPT Model Support: Token tracking and model routing updated for the latest GPT-5.x deployments.
- Evaluation API: New endpoint for automated evaluation runs.
Fix for Context Window Overflow in the Response Endpoint (v1.20.0)
Bugfixes and Improvements:
- The response endpoint now returns an error message when the context window is exceeded
Support for New Models and Enhancements for Reports, Pricing, and Token Limits (v1.20.0)
Features:
- Support for new models added: gpt-5.4 and gpt-5.5
- Reports can now be exported as JSON or CSV
- Tiered pricing based on input tokens is now supported
- Input tokens can now be limited separately per deployment for frontend and API
Bugfixes and Improvements:
- Image models now return an error message if generation fails due to the content filter
- The response endpoint now returns an error message when the context window is exceeded
ML compute quota now shown in GPU-h
We would like to inform users of the following changes to r_wlm_usage.
Users can now see that the r_wlm_usage and r_wlm_usage -q commands display different quotas for MPI (CPU Only) and ML (GPUs) projects. The r_wlm_usage table will display for the same project, two distinct sections as follows:
=== HPC partitions (CPU Core-h) ===
….
=== ML partitions (GPU-h) ===
….
These changes are mostly in how the values are displayed as GPU-h. The used quotas have not changed and we expect no negative effects to previous, current or future quota utilization. Already used and allocated quotas within ML (GPU) systems were converted with the pre-existing ratio of 24CPU-h per 1 GPU-h. This change is to help distinguish between CPU and GPU usage within the NHR directives.
Additionally the default GPU-h quota for the ‘default’ project of every account running on GPU nodes has been set to 10 GPU-h.
This means that users can use their ‘default’ CPU-h quota without using the GPU-h quota and vice versa. The same rules of over-quota apply to both types.
