Announcements
Major launches, new models, and platform updates from the Nova team.
Nova Fusion is now in beta — blend any two models into one response
Nova Fusion lets you specify weights across multiple models and receive a single synthesized response. Early benchmarks show 12–18% accuracy improvements over single-model calls on complex reasoning tasks.
Video generation is live — Kling, Wan, Hailuo, and Seedance in one API
Four leading video generation models are now available through Nova's unified endpoint. Generate up to 60-second clips at 720p with a single API call and OpenAI-compatible parameters.
Introducing Workspaces — separate environments for every project
Workspaces give teams isolated API keys, usage budgets, and observability per project. Manage multiple products, clients, or environments from a single Nova account.
State of AI 2026: analysis of 18 trillion tokens on Nova
We analyzed 18 trillion tokens of anonymized API traffic to understand how developers actually use AI models. The findings on task distribution, model switching, and geographic growth surprised us.
DeepSeek R1 Zero, Qwen 3 235B, and Kimi K2 now available
Three frontier open-weight models added in one week. DeepSeek R1 Zero leads MATH-500 benchmarks at 95.9%. All three are available at competitive prices with no waitlist.
Global edge routing — p50 latency down 38% across all regions
We rebuilt our request routing layer across five regions. Requests are now dispatched to the nearest available provider replica, cutting median latency from 290ms to 180ms globally.
Audio models: ElevenLabs, Whisper, and OpenAI TTS now on Nova
Text-to-speech and speech-to-text are now part of the Nova API. ElevenLabs Flash v2.5, Whisper Large v3, and OpenAI TTS-1 are live with the same OpenAI-compatible request format.
Nova Enterprise — SSO, SLAs, and volume pricing for teams
Enterprise plan now available with SAML 2.0 / Okta SSO, 99.9% uptime SLA, dedicated support, volume discounts, and data privacy guarantees. Contact sales to get started.
Nova partners with a16z on AI infrastructure research
Andreessen Horowitz will use Nova's anonymized usage data to study enterprise AI adoption patterns. This builds on our ongoing data transparency program and the State of AI report.
Batch API preview — process 100k requests at 50% lower cost
Early access to async batch processing is now open. Submit a JSONL file, get results back within 24 hours, and pay half the per-token rate. Ideal for evals, dataset generation, and bulk tasks.