SID AI launches SID-2 agentic search model and 12-agent Swarm
SID AI has released SID-2, an agentic search model designed to query data, read returned results and produce ranked document IDs. Unlike a fixed retrieval-augmented generation pipeline, it can decide what to search next. Developers connect its text_search and search tools to their own keyword and vector-search systems, without reindexing or moving data. The model has a reported 262k-token context window, while SID’s website lists 256k.
SID-2 was trained with two-stage reinforcement learning, including a latency penalty for positive rewards. It offers low, medium and high reasoning effort, and Metric Steering for NDCG, F1 or Recall@k. On SID’s self-reported 506-question benchmark covering 19 corpora, SID-2 Medium reached 0.585 NDCG in 3.11 seconds. The six-agent Swarm scored 0.701 in 17.7 seconds, while the 12-agent version reached 0.720, compared with 0.704 for GPT-6 Astra.
The model is available through SID’s API, Baseten and SageMaker reference paths. Pricing starts at $0.25 per million input tokens and $1.00 per million output tokens; Swarm costs $0.50 and $2.00. SID has not disclosed model size or weights, and the reported evaluations have not been independently verified.
Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.