Release 12.3 | Clarifai Docs

Skip to main content

GPU Shortage?Make the most out of your GPUs with Compute Orchestration. Learn More

On this page

Release Date: April 8th, 2026

New Feature Improvement Bug Fix Enterprise Only

Agent Skills ​

Status Change Details
Released Clarifai Skills - Clarifai Skills are specialized prompt templates that transform AI coding assistants — Claude Code, Cursor, Codex, and more — into Clarifai platform experts.
- Learn more about them here.

Model Training ​

Status Change Details
Added model pipeline training from templates - You can now train models using pipeline templates, enabling a streamlined, configuration-driven training workflow.
Deprecation Deprecated legacy model training (Triton + Kubeflow) - Legacy model training methods using Triton and Kubeflow have been deprecated.
- Note: Transfer Learn training remains available.

Request Routing ​

Status Change Details
Improved how Clarifai routes prediction requests for optimal performance - Added KV cache affinity to route requests to replicas with relevant cache state.
- Added session-aware routing to keep user requests on the same replica.
- Reduced cold starts with automatic pre-warming of popular instances.
- Added prediction caching for identical input + model + version combinations.
- Learn more about them here.

UI Updates ​

Status Change Details
Updated Compute List and View pages - Refreshed the List and View pages for deployments, nodepools, and clusters with improved layouts and information display.
Updated the log viewer component - Improved the log viewer UI for better readability and navigation of model and runner logs.
Updated the Home experience - Refreshed the Clarifai platform Home page with an improved experience for navigating resources and getting started.
Updated the model page UI - Redesigned model page with an improved layout and user experience.

Python SDK ​

Model Serving & Deployment ​

Status Change Details
Added clarifai model deploy command and simplified clarifai model init - New clarifai model deploy command with multi-cloud GPU discovery and a zero-prompt deployment flow.
- Simplified config.yaml structure for model initialization.
Smart resource reuse and private-by-default for clarifai model serve - Model serve now reuses existing resources when available instead of creating new ones.
- Served models are private by default.
Added --keep flag to clarifai model serve - Use --keep to preserve the build directory after serving, useful for debugging and inspecting build artifacts.
Local Runner is now public by default - Models launched via the local runner are now publicly accessible by default, removing the need to manually set visibility.

Model Runner ​

Status Change Details
Added VLLMOpenAIModelClass - New VLLMOpenAIModelClass parent class with built-in cancellation support and health probes for vLLM-backed models.
Optimized model runner memory and latency - Reduced memory footprint and improved response latency in the model runner.
- Streamlined overhead in SSE (Server-Sent Events) streaming.
Auto-detect and clamp max_tokens - The runner now automatically detects the backend's max_seq_len and clamps max_tokens to that value, preventing out-of-range errors.

Bug Fixes ​

Status Change Details
Fixed reasoning model token tracking and streaming in agentic class - Fixed token tracking for reasoning models to correctly account for reasoning tokens.
- Fixed event-loop safety, streaming, and tool call passthrough in the agentic class.
Fixed user/app context conflicts in CLI - Resolved conflicts between user_id and app_id when using named contexts in CLI commands.
Fixed clarifai model init directory handling - clarifai model init now correctly updates an existing model directory instead of creating a subdirectory.