This blog post focuses on new features and improvements. For a comprehensive list, including bug fixes, please see the release notes.
Model deployment on Clarifai is now faster and easier. Previously, users had to manually configure clusters and nodepools before deploying a model, with limited setup guidance.
With Single-Click Deployment, Clarifai now recommends suitable instance types based on each model’s requirements and automatically creates clusters or nodepools if none exist. This removes the need for any manual setup, allowing users to deploy models instantly.
The platform intelligently matches compute resources to model needs, ensuring the right GPU type, memory, and core allocation for every deployment. For Premium GPUs such as the NVIDIA B200, users can reach out through the integrated Contact Us option to provision dedicated instances for higher performance.
This update eliminates unnecessary steps, reduces setup errors, and makes production deployment possible in a single click. Check out the complete guide here on the Custom Model Deployment Guide.
DeepSeek-OCR sets a new standard for large-scale document understanding and OCR performance. It delivers over 96% precision at 9–10× compression, and around 90% accuracy even at 10–12× compression, maintaining reliability under heavy optimization.
Designed for production-grade scalability, DeepSeek-OCR can process over 200,000 pages per day on a single A100-40G GPU, enabling enterprise-level document automation at a fraction of typical compute cost.
You can try DeepSeek-OCR directly in the Playground or access it through the API. Check out the detailed DeepSeek-OCR API Guide.
The GLM-4.6 model brings together reasoning, code understanding, and agentic capabilities into a single unified framework. It’s optimized for multi-domain tasks where models need to analyze, plan, and generate in a structured manner.
GLM-4.6 enables consistent reasoning performance across natural language, programming, and tool-using contexts, making it ideal for developers building intelligent agents or multi-skill assistants.Try out the model here.
The Control Center now provides a single, consistent view of model usage across all billing methods.
Previously, usage statistics were tied to the billing configuration. Ops-billed models reported only operations, token-billed models reported only tokens, and models billed by compute time did not display detailed stats.
With this update, all models now report operations, and LLMs additionally report token usage. This ensures consistent visibility and transparent tracking for every model, regardless of how it’s billed.
The result is a more reliable and unified monitoring experience for developers and teams managing large-scale deployments.
Clarifai now supports structured JSON outputs from any OpenAI-compatible model hosted on the platform using Pydantic schemas.
This capability ensures that model responses follow a defined schema, allowing developers to enforce consistent data structures across outputs. Structured outputs make it easier to integrate AI-generated data into downstream applications safely and reliably.
Here’s an example using the GPT-OSS-120B model through Clarifai’s OpenAI-compatible API:
The Community search experience has been refined to surface more relevant results.
Previously, all fields such as model ID, user ID, and description were weighted equally in search ranking. With this update, model IDs (for example, gpt-oss-120b) now carry higher weight, ensuring that searches prioritize the most relevant and specific models.
Clarifai now supports environment secrets, allowing developers to securely store encrypted values that can be referenced as environment variables in workflows.
This improves security and simplifies management of credentials and other sensitive configuration data. Learn more about environment secrets here.
Support for additional toolkits has been added to the Clarifai CLI, making it easier to initialize model projects with pre-configured templates.
Developers can now specify a toolkit when creating a new model project using the clarifai model init command:
SGLang Toolkit: clarifai model init --toolkit sglang
Python Toolkit: clarifai model init --toolkit python
These toolkits streamline setup, ensuring consistency and faster onboarding for both SGLang-based and Python-based model development. Check out the detailed Toolkit Guide here.
With Single-Click Deployment, Clarifai makes it easier than ever to bring your own models and deploy them in production with minimal setup. The platform automatically manages cluster creation, instance selection, and scaling, allowing you to focus on iterating and improving your models instead of configuring infrastructure.
You can start by deploying your own model using the new one-click workflow or explore the growing catalog of community and published models.
If you need access to high-end GPUs like the B200 or GH200 for your AI workloads, reach out to our team to learn more about dedicated provisioning and performance optimization options.
