voyage-code-4: code retrieval built for coding agents
Voyage AI has launched voyage-code-4, a next-generation code embedding model designed for coding agents. The model is priced at $0.12 per 1M tokens and outperforms voyage-code-3 on agentic code retrieval benchmarks.
Also shipped Aug 13, 2026Voyage AI in August 2026 →
BigQuery pipelines add automated metadata enrichment in preview
Semantic metadata defined in SQLX configurations now syncs with Knowledge Catalog, and the Data Engineering Agent can generate this metadata from your intent or context.
Sharing tables backed by default storage will be enabled by default for all accounts
OpenSharing support for tables backed by default storage is generally available and on by default for most accounts. For most Enterprise or Premium tier accounts, Databricks will enable it automatically in mid-September 2026, and by early 2027 for the rest.
Expanded networking billing for OpenSharing SecureConnect
Databricks will be expanding networking billing for OpenSharing SecureConnect. Today, data providers are billed for data transfer (egress) when recipients access shared data through SecureConnect. Additional charge types and access scenarios, including charges billed to recipients, are planned.
OpenSharing SecureConnect will be available by default for all accounts
OpenSharing SecureConnect is generally available and available by default for accounts on the Premium plan. For accounts on the Enterprise plan, Databricks will enable it automatically in mid-September 2026, and by early 2027 for the rest.
Block or redact sensitive data in requests and responses
The Sensitive Data Detection guardrail (Beta) detects structured sensitive data (such as Social Security numbers, credit card numbers, IBANs, and phone numbers) and either blocks the interaction or redacts the matched values in place.
Smart Routing: Cut AI costs by routing each task to the right model
Smart Routing (Beta) automatically picks the right model and agent harness for each task, based on complexity, permissions, and budget. It shifts routine work to lower-cost models and reserves frontier models for hard problems, and every routing decision is visible and auditable.
More from Voyage AIEvery Voyage AI release →
voyage-context-4: stop worrying about chunking with our best-performing model
Voyage AI has released voyage-context-4, a contextualized chunk embedding model that features auto-chunking and handles documents longer than 32K tokens.
rerank-3 and rerank-3-lite: the next generation of Voyage rerankers
Voyage AI introduced the Rerank 3 series, offering improved retrieval performance on long documents and code compared to previous versions. The models maintain a 32K token context length and instruction-following capabilities.
Announcing New Models and Expanded Availability
Voyage AI introduced the Voyage 4 embedding model series with shared embedding spaces and the voyage-multimodal-3.5 model with video retrieval support. The models are now available on MongoDB Atlas and GCP.
The Voyage 4 model family: shared embedding space with MoE architecture
The Voyage 4 series features a shared embedding space for cross-model compatibility. The flagship voyage-4-large model uses a mixture-of-experts architecture to improve retrieval accuracy while reducing costs.
voyage-multimodal-3.5: a new multimodal retrieval frontier with video support
The voyage-multimodal-3.5 model adds support for embedding video frames alongside text and images. It also supports Matryoshka embeddings for flexible dimensionality.
Introducing the Batch API: Simpler and more efficient for large-scale workloads
The new Batch API is an asynchronous endpoint for processing large volumes of requests. It offers higher throughput, a 12-hour completion window, and 33% cost savings compared to synchronous alternatives.