New Models Added
- Added Microsoft Phi 4 (16K context)
- Added DeepSeek R1 Qwen3 8B (32K context)
- Reduced inference latency by 20% across all serverless endpoints
- Improved vector search performance for large indexes
New Integrations
- Added PydanticAI support for type-safe AI applications
- Enhanced n8n workflow automation capabilities
- Extended dimension support up to 2000 dimensions
- Added dot product similarity metric option
Vision Models Launch
- Introduced Google Gemma 3 12B with vision capabilities
- Added Qwen 2.5 VL 7B for multimodal tasks
- Improved structured outputs with schema validation
- Enhanced function calling reliability
Dedicated Deployment Improvements
- Added horizontal scaling support
- Implemented auto-scaling for GPU instances
- Enhanced resource isolation for enterprise workloads
Extended Context Windows
- Llama 3.1 and 3.2 models now support up to 128K context
- Mistral Nemo 12B upgraded to 128K context window
- Increased file upload limit to 200MB
- Added support for additional file formats
Platform Launch
- Public preview launch of Gravix Layer
- OpenAI-compatible API endpoints
- Serverless inference for immediate deployment
- Chat completions with system prompts
- Embedding generation for RAG applications
- Vector database with similarity search
- File upload and management system
Beta Release
- LangChain integration testing
- CrewAI multi-agent framework support
- Initial MongoDB and Pinecone connectors
Model Testing Phase
- Meta Llama 3.1 8B integration
- Mistral 7B Instruct deployment
- Qwen3 14B performance optimization
- Basic vector operations (upsert, search, delete)
- Cosine and euclidean similarity metrics