SID AI Releases SID-2 Agentic Search Model With 0.720 NDCG Swarm Mode

What happened
SID AI introduced SID-2, its second-generation agentic search model, along with SID-2 Swarm, on its hosted API and partner platforms. The new model searches and ranks documents by dynamically deciding what to search next, rather than following a fixed pipeline. In internal benchmarks, SID-2 Swarm, using 12 agents, achieved a 0.720 NDCG score, surpassing GPT-6 Astra. The model is available through SID’s API and supports large context windows and configurable reasoning levels.
Key details
- SID-2 Swarm (12 agents) scored 0.720 NDCG, above GPT-6 Astra’s 0.704 on SID’s benchmark.
- Context window is 262,000 tokens (report) or 256,000 tokens (site).
- SID-2 Swarm costs $0.50 per million input tokens and $2.00 per million output tokens.
- Developers can access SID-2 via API, Baseten, SageMaker, and SDKs for Python, TypeScript, and Go.
- SID-2 offers metric steering for NDCG, F1, or Recall@k, and three levels of reasoning effort.
Why it matters
Developers and companies needing high-quality document retrieval now have access to a model that, according to SID AI’s benchmarks, outperforms leading alternatives at a lower cost and with faster response times. Organizations can deploy SID-2 without reindexing their data and can adjust retrieval quality and cost by configuring reasoning effort and metric targets. This release gives users more control over search performance and pricing, potentially reducing operational expenses for large-scale document search tasks.
