LLM Studio Release Notes¶

This page lists the release notes for LLM Studio, helping you understand version evolution and feature changes.

2026-08-31¶

v0.18.3¶

User View

• Added support for administrators to assign created API Keys to users, so that users can use the API Keys assigned to them
• Added support for viewing and displaying the usage of deleted API keys in usage statistics
• Added a workload viewing action in the model deployment list, so that users can quickly jump to the details of the associated workload
• Added support for displaying multimodal model API call examples in the model gallery details

Admin Console

• Added intelligent routing, where each cluster can be configured with one model=auto route. It supports basic rule-based routing (request length and custom rules) and advanced semantic model selection (domain classification, semantic embedding, complexity, and context length), and falls back to the default model when no rule matches
• Added support for token-based rate limiting for MaaS models
• Added content security detection policy management, which calls security capabilities to perform risk and compliance detection on request content. General security detection is completed by the platform's own security model, with configurable risk types and risk determination thresholds. Module-shield security detection is an integration with the NSFOCUS AI security fence service, completing content detection through NSFOCUS open APIs

2026-07-31¶

v0.17.1¶

User View

• Added support for image-to-image generation and voice model experiences in Model Playground
• Added support for environment variable configuration for fine-tuning and evaluation tasks
• Optimized the Model Marketplace interface, including display, filtering, and sorting capabilities, to improve the user experience
• Optimized distributed inference scenarios for model deployment, allowing separate startup commands and environment variables to be configured for Leader and Worker nodes

2026-06-30¶

v0.16.0¶

User View

• Added container health checks for model deployments, including liveness, readiness, and startup probes
• Added support for enabling metrics collection for model deployments to monitor their runtime status
• Added support for mounting external configuration files to customize the runtime logic of model deployments

2026-05-31¶

v0.15.0¶

User View

• Added support for large language model fine-tuning, including dataset management, fine-tuning job management, model evaluation, and model export
• Added pre-deployment health checks for model deployments, automatically validating critical hardware and software environments before deployment to identify issues in advance
• Added support for token weight configuration in model deployments and MaaS models to ensure fair token quota allocation
• Added support for enabling and disabling API Keys, allowing users to activate API Keys as needed
• Added support for accounting of cached tokens in large language models
• Added support for workspace token quota configuration in operations management, allowing users to set token limits for each workspace

Admin Console

• Added support for gateway security policy management and audit logs in operations management, allowing users to configure gateway security policies such as sensitive word detection and model allowlists/blocklists
• Added support for forwarding multimodal MaaS models, including speech-to-text, OCR, rerank, and embedding models
• Improved model marketplace management by adding support for model deletion

2026-04-30¶

v0.14.0¶

User View

• Added support for API Key expiration settings
• Added support for API Key token quota settings
• Improved API Key management permissions, allowing each user to create and manage their own API Keys
• Improved model experience authentication to support only user-created API Keys
• Improved model deletion logic by adding validation for associated model services
• Improved model deployment templates by supporting custom GPU types and custom runtime logic for more flexible model deployment

2026-03-31¶

v0.13.1¶

User View

• Added support for user private model management, including configuration of deployment templates and management of model weight files
• Added support for user-defined deployments, allowing modification of resource and runtime configurations during deployment
• Added support for priority and topology-aware queue scheduling in inference services
• Added support for user-defined runtime frameworks in inference services
• Added support for creating and managing file storage for persistent storage of model files, datasets, etc.

Admin Console

• Added file storage module for uploading and persistent storage of model weight files in the model hub
• Added UI-based management of model weight files in the model hub
• Optimized deployment configuration management by integrating it into model hub management, supporting multiple configuration templates

2025-11-30¶

v0.12.1¶

Admin Console

• Added support for custom inference frameworks in model deployment management.

2025-10-31¶

v0.11.0¶

Admin Console

• Added support for uploading model icons locally in the Model Marketplace.

2025-09-30¶

v0.10.0¶

Admin Console

• Added support for deploying models using SGLang.
• Added support for viewing audit logs.

2025-08-07¶

v0.8.0¶

Admin Console

• Added model list display
• Added support for bulk importing models from URL address information
• Added support for creating models
• Added MaaS model list display
• Added support for adding MaaS models
• Added support for adding multiple upstream endpoint configurations
• Added support for rate limiting at the apikey level
• Added support for rate limiting at the Workspace level
• Added support for defining multiple sets of rate-limiting rules
• Added support for round-robin load balancing
• Added model deployment management list display
• Added support for configuring multiple sets of model parameters
• Added support for deploying models with vLLM

2025-07-04¶

v0.7.0¶

User View

• Added support for deep thinking in text models.
• Added support for copying and regenerating messages in text models.
• Added support for generating multiple images in vision-to-text models.
• Added support for custom positive and negative prompts, as well as custom image sizes in vision-to-text models.
• Added full compatibility with the standard OpenAI SDK.
• Added support for usage statistics by API Key, model type, and call time.
• Added quick summaries for total invocations, input tokens, and output tokens.
• Added support for usage comparison across multiple models.
• Added card view for displaying model lists with visual summaries.
• Added support for showing model details and API call examples.
• Added support for quick deployment and trial of text models.
• Added support for searching models by name, provider, and type.
• Added support for trialing multimodal models.
• Added support for trialing vision-to-text models.
• Added support for listing deployed model services.
• Added support for filtering models by type for quick deployment.
• Added support for scaling model service instances horizontally.
• Added support for configuring the number of model service instances.
• Added support for online trials of deployed model services.
• Added API call examples in curl, Python, and Node.js for deployed models.
• Added one-click trial feature for quickly verifying service availability.
• Added support for configuring parameters for text generation models, including system prompt, temperature, and top_p.
• Added support for comparison among models of the same type.
• Added support for displaying API Key lists.
• Added one-click copy for full API Keys.
• Added support for deleting specific API Keys irreversibly.
• Added support for generating new API Keys with custom names.

Comments