This weblog publish focuses on new options and enhancements. For a complete record, together with bug fixes, please see the launch notes.
Single-Click on Deployment
Mannequin deployment on Clarifai is now sooner and simpler. Beforehand, customers needed to manually configure clusters and nodepools earlier than deploying a mannequin, with restricted setup steering.
With Single-Click on Deployment, Clarifai now recommends appropriate occasion varieties primarily based on every mannequin’s necessities and robotically creates clusters or nodepools if none exist. This removes the necessity for any guide setup, permitting customers to deploy fashions immediately.
The platform intelligently matches compute assets to mannequin wants, guaranteeing the proper GPU sort, reminiscence, and core allocation for each deployment. For Premium GPUs such because the NVIDIA B200, customers can attain out by means of the built-in Contact Us choice to provision devoted cases for increased efficiency.
This replace eliminates pointless steps, reduces setup errors, and makes manufacturing deployment attainable in a single click on. Take a look at the whole information right here on the Customized Mannequin Deployment Information.
New Fashions
DeepSeek-OCR: Excessive-Precision Textual content Extraction at Scale
DeepSeek-OCR units a brand new normal for large-scale doc understanding and OCR efficiency. It delivers over 96% precision at 9–10× compression, and round 90% accuracy even at 10–12× compression, sustaining reliability underneath heavy optimization.
Designed for production-grade scalability, DeepSeek-OCR can course of over 200,000 pages per day on a single A100-40G GPU, enabling enterprise-level doc automation at a fraction of typical compute price.
You’ll be able to strive DeepSeek-OCR straight within the Playground or entry it by means of the API. Take a look at the detailed DeepSeek-OCR API Information.
GLM-4.6: Unified Reasoning, Coding, and Agentic Intelligence
The GLM-4.6 mannequin brings collectively reasoning, code understanding, and agentic capabilities right into a single unified framework. It’s optimized for multi-domain duties the place fashions want to research, plan, and generate in a structured method.
GLM-4.6 permits constant reasoning efficiency throughout pure language, programming, and tool-using contexts, making it very best for builders constructing clever brokers or multi-skill assistants.Check out the mannequin right here.
Management Middle: Unified Ops and Token Reporting
The Management Middle now supplies a single, constant view of mannequin utilization throughout all billing strategies.
Beforehand, utilization statistics had been tied to the billing configuration. Ops-billed fashions reported solely operations, token-billed fashions reported solely tokens, and fashions billed by compute time didn’t show detailed stats.
With this replace, all fashions now report operations, and LLMs moreover report token utilization. This ensures constant visibility and clear monitoring for each mannequin, no matter the way it’s billed.
The result’s a extra dependable and unified monitoring expertise for builders and groups managing large-scale deployments.
Structured Outputs
Clarifai now helps structured JSON outputs from any OpenAI-compatible mannequin hosted on the platform utilizing Pydantic schemas.
This functionality ensures that mannequin responses observe an outlined schema, permitting builders to implement constant information buildings throughout outputs. Structured outputs make it simpler to combine AI-generated information into downstream purposes safely and reliably.
Right here’s an instance utilizing the GPT-OSS-120B mannequin by means of Clarifai’s OpenAI-compatible API:
Extra Adjustments
Search by Relevance in Neighborhood
The Neighborhood search expertise has been refined to floor extra related outcomes.
Beforehand, all fields resembling mannequin ID, person ID, and outline had been weighted equally in search rating. With this replace, mannequin IDs (for instance, gpt-oss-120b) now carry increased weight, guaranteeing that searches prioritize probably the most related and particular fashions.
Setting Secrets and techniques
Clarifai now helps atmosphere secrets and techniques, permitting builders to securely retailer encrypted values that may be referenced as atmosphere variables in workflows.
This improves safety and simplifies administration of credentials and different delicate configuration information. Study extra about atmosphere secrets and techniques right here.
Toolkits
Help for extra toolkits has been added to the Clarifai CLI, making it simpler to initialize mannequin tasks with pre-configured templates.
Builders can now specify a toolkit when creating a brand new mannequin venture utilizing the clarifai mannequin init command:
These toolkits streamline setup, guaranteeing consistency and sooner onboarding for each SGLang-based and Python-based mannequin growth. Take a look at the detailed Toolkit Information right here.
Able to Begin Constructing?
With Single-Click on Deployment, Clarifai makes it simpler than ever to carry your personal fashions and deploy them in manufacturing with minimal setup. The platform robotically manages cluster creation, occasion choice, and scaling, permitting you to deal with iterating and bettering your fashions as a substitute of configuring infrastructure.
You can begin by deploying your personal mannequin utilizing the brand new one-click workflow or discover the rising catalog of neighborhood and revealed fashions.
In the event you want entry to high-end GPUs just like the B200 or GH200 on your AI workloads, attain out to our group to be taught extra about devoted provisioning and efficiency optimization choices.
