SID AI has released SID-2, a second-generation agentic search model, along with SID-2 Swarm, MarkTechPost reports. SID-2 is not a general chat model. It searches your data, reads the results, decides what to look at next, and returns a ranked list of document IDs, instead of running a fixed rewrite-embed-search-rerank pipeline. Developers implement the search tools against their own backend, including keyword and vector search. The model only sees what those tools return, so SID says you do not have to reindex or move the corpus. SID-2 Swarm runs several agents that can message each other and fork their context, with each agent exploring a different slice of the collection.
SID reports a 262k-token context window in the technical report (256k on its models page), a hosted API at platform.sid.ai, and reference paths on Baseten and SageMaker. The benchmarks in the coverage are SID’s own. A single SID-2 agent is said to match Claude Opus 5.5 on retrieval quality, and a 12-agent swarm is reported at 0.720 NDCG, just above GPT-6 Astra at 0.704, on SID’s 506-question set. Astra still leads on some of those corpora, and SID-2 Low scores well behind. SID has not disclosed model size or weights, and there are no independent evaluations in the write-up I saw.
My take
An agent that just returns ranked document IDs, against your own search tools, is a cleaner shape than another all-knowing chatbot. I have not used SID-2, and I am treating the scores as self-reported until someone else reruns them. If you already have a search backend, the API is the thing to look at, not a local download that does not exist yet.
God Bless, and Use the Source!