Nova Labs

Blend outputs from multiple models in a single API call. Specify weights per model and Fusion synthesizes their responses into one coherent answer — better accuracy than any single model alone.

• ✓Weighted model blending
• ✓Parallel inference
• ✓Consensus scoring
• ✓OpenAI-compatible endpoint

Process thousands of requests asynchronously at 50% lower cost. Submit a JSONL file, get results back within 24 hours. Ideal for dataset generation, evaluation pipelines, and bulk summarization.

• ✓Up to 50% price reduction
• ✓100k requests per batch
• ✓JSONL input/output
• ✓Webhook notifications

Deploy models inside your own VPC with zero data leaving your network. Full model weights, no shared infrastructure, SOC 2 + HIPAA compliant. Built for regulated industries and high-sensitivity workloads.

• ✓VPC deployment
• ✓Zero egress
• ✓HIPAA & SOC 2
• ✓Dedicated GPU allocation

Define routing rules to automatically send requests to the right model based on task type, cost budget, or latency requirements — without changing your code. Uses a small classifier to categorize each request.

• ✓Rule-based routing
• ✓Cost-aware selection
• ✓Latency targets
• ✓Fallback chains

Run automated evaluations across models on your own test cases. Score outputs by accuracy, tone, format, and custom rubrics. Compare models side-by-side and track quality over time as models update.

• ✓Custom rubrics
• ✓A/B model comparison
• ✓Human-in-the-loop review
• ✓Regression tracking

Fine-tune open-weight models on your own data directly through the Nova API. Upload training data, monitor loss curves, and deploy custom endpoints — all without managing GPU infrastructure.

• ✓LoRA / QLoRA adapters
• ✓Custom base models
• ✓One-click deployment
• ✓Per-request routing