Nace AI Launches Drex for Search: 92.05% on BrowseComp-Plus with 12.6 Search Calls per Question
Nace AI has released Drex for Search, extending its Drex decision models from agentic reasoning into document search. Drex for Search is a hosted retrieval layer designed to help both people and AI agents find evidence efficiently. Coding agents such as Claude Code, Codex, Cursor, and OpenCode can call it through a CLI, SDKs, MCP, or REST.
In Nace's own evaluation, Drex for Search scores 92.05% on BrowseComp-Plus with GPT-5, using just 12.6 search calls per question β a notable efficiency result as agentic search benchmarks in 2026 increasingly reward accuracy per retrieval step rather than raw call volume.
TL;DR
- Scale: A hosted search system built to search over 1M+ files. It pairs with an answering model such as GPT-5, Gemini 3.5 Flash, or Qwen 27B.
- Deployment: Runs on Nace's managed cloud, a dedicated cloud, your own AWS or Azure account, or on-prem and air-gapped environments.
- Performance: Ranked #1 or #2 on 3 of 4 public leaderboards, according to Nace.
- Integration: Accessible via CLI, SDKs, MCP, or REST for coding agents and enterprise workflows.
- Positioning: A retrieval layer that separates evidence-finding from answer generation, allowing teams to swap answering models without rebuilding search infrastructure.
Why It Matters
As AI agents take on more autonomous research and coding tasks in 2026, the bottleneck has shifted from model capability to reliable evidence retrieval. Drex for Search targets this gap by decoupling retrieval from generation: agents query the layer for evidence, while a separate answering model synthesizes the response. This architecture lets organizations pair Drex with whichever frontier model fits their cost, latency, or compliance requirements β including GPT-5, Gemini 3.5 Flash, or Qwen 27B.
Deployment and Integration
Drex for Search is available across a range of deployment options to suit different security and infrastructure needs:
- Nace's managed cloud
- A dedicated cloud instance
- Your own AWS or Azure account
- On-premises and air-gapped environments
The system integrates with popular coding agents through a CLI, SDKs, the Model Context Protocol (MCP), or REST APIs, making it accessible to both developer tooling and enterprise agent platforms.
Performance Snapshot
| Metric | Result |
|---|---|
| BrowseComp-Plus score (with GPT-5) | 92.05% |
| Search calls per question | 12.6 |
| Public leaderboard rankings | #1 or #2 on 3 of 4 leaderboards (per Nace) |
| Supported corpus size | 1M+ files |
Looking Ahead
With agentic search benchmarks maturing in 2026, the focus is shifting toward efficient, verifiable retrieval. Nace's claims around accuracy-per-call and multi-cloud deployment flexibility position Drex for Search as a candidate for enterprise agent stacks that require both performance and data sovereignty.
via MarkTechPost