🧠Weaviate Blog•Freshcollected in 15h
Weaviate Adds Effort Tiers to Search Mode
💡Tune agentic search quality and cost with Weaviate's new test-time compute tiers.
⚡ 30-Second TL;DR
What Changed
Query Agent's Search Mode now supports medium, high, and ultrahigh effort tiers.
Why It Matters
This gives developers finer control over the quality–latency–cost balance of agentic search. Higher-effort modes may be useful for complex retrieval tasks, while lower tiers can help control resource usage for routine queries.
What To Do Next
Test the three Query Agent Search Mode effort tiers on a representative query set and compare answer quality, latency, and compute cost.
Who should care:Developers & AI Engineers
Key Points
- •Query Agent's Search Mode now supports medium, high, and ultrahigh effort tiers.
- •The tiers provide a way to scale test-time compute during search.
- •Practitioners can choose different effort levels based on quality, latency, and compute trade-offs.
🧠 Deep Insight
AI-generated analysis for this event.
🔑 Enhanced Key Takeaways
- •The effort tiers utilize test-time compute scaling, a technique often associated with chain-of-thought reasoning or iterative refinement processes to improve retrieval accuracy.
- •This feature is specifically integrated into Weaviate's Query Agent, which acts as an autonomous orchestration layer for managing complex RAG (Retrieval-Augmented Generation) workflows.
- •The 'Ultrahigh' tier is designed for complex, multi-hop reasoning tasks where standard vector search might fail to capture nuanced semantic relationships.
- •Implementation of these tiers allows developers to dynamically adjust cost-per-query, enabling cost-optimization for simple queries while reserving expensive compute for high-stakes retrieval.
- •The update reflects a broader industry shift toward 'inference-time compute' where models are given more time and resources to process a single request to improve output quality.
📊 Competitor Analysis▸ Show
| Feature | Weaviate (Query Agent) | Pinecone (Serverless/Assistant) | Milvus (Managed) |
|---|---|---|---|
| Test-Time Compute Scaling | Native Effort Tiers | Limited/Not Explicit | Via Custom Pipelines |
| Orchestration | Built-in Query Agent | Via External Frameworks | Via External Frameworks |
| Pricing Model | Tiered Compute/Usage | Usage-based | Usage-based |
🛠️ Technical Deep Dive
- The effort tiers manipulate the depth of the Query Agent's reasoning loop, likely increasing the number of iterations or the complexity of the sub-queries generated during the retrieval phase.
- Integration leverages Weaviate's existing vector search engine as the base retrieval mechanism, with the Query Agent acting as a meta-layer that performs query expansion, decomposition, or re-ranking.
- The system likely utilizes a cost-aware scheduler that maps the selected effort tier to specific resource allocation parameters within the Weaviate cloud infrastructure.
- The architecture supports dynamic prompt engineering or tool-use cycles that scale linearly or exponentially based on the selected tier (Medium vs. Ultrahigh).
🔮 Future ImplicationsAI analysis grounded in cited sources
Standard vector search will become a commodity while 'reasoning-based' retrieval becomes the primary value driver.
The shift toward test-time compute suggests that the industry is moving away from simple similarity search toward complex, agentic retrieval workflows.
Developers will increasingly adopt 'tiered-latency' architectures for AI applications.
By offering variable effort levels, Weaviate is enabling a new design pattern where application latency is adjusted dynamically based on the user's specific query complexity.
⏳ Timeline
2023-05
Weaviate secures $50M Series B funding to accelerate vector database development.
2024-02
Weaviate introduces the 'Verba' RAG application to simplify retrieval-augmented generation.
2025-01
Weaviate launches the Query Agent to enable autonomous, multi-step search capabilities.
2026-08
Weaviate adds effort tiers to Query Agent's Search Mode to scale test-time compute.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Weaviate Blog ↗