How SID AI launched SID-2 on X
Search model whose compute-scaling swarms return better results faster and cheaper
What SID AI launched

Introducing SID-2, a new search model built on compute-scaling swarms. Our swarms find better results than Opus and Astra, while being 8x faster and 20x lower cost. By 2030, AI will write an internet's worth of text every 24 hours. Current search pipelines were built on https://t.co/OGGUyMeDu2
Transcript
How LaunchVideo.com rated the launch
LaunchVideo.com rating
About the launch
SID AI has released SID-2, a search model for private data, and SID-2 Swarm, which runs that model as a coordinated team. It takes a natural-language question, queries a customer's existing search tools, reads the results, and returns a ranked list of relevant documents rather than a written answer. Teams can use it as the retrieval step for agents, support, recruiting, commerce, and research over legal, finance, and other internal corpora, without moving data or reindexing. The release tracks a volume problem in SID's technical report: as of November 2026, the lab estimates that language models already generate as many output tokens as exist on the public web every 15 days, while BM25, embedding, and reranker pipelines still spend a fixed amount of compute on every query. SID-2 is trained with reinforcement learning. SID starts from a small agent that can search in about 300 milliseconds and scales it to swarms of up to 24 agents that fork into separate contexts, split the work, and share findings. On the lab's benchmark of hard questions, medium-effort SID-2 reaches 0.58 NDCG in 3.1 seconds of model inference, matching what Claude Opus 5.5 at medium effort achieves in 25 seconds. Six agents reach 0.70 NDCG in 18 seconds, matching what GPT-6 Astra at maximum effort achieves in 135 seconds, and a 12-agent swarm posts the highest aggregate NDCG in the report, ahead of Astra. Astra still leads on several legal and financial corpora, at much higher latency and cost. SID says the swarms are 8 times faster and 20 times more cost efficient than Opus and Astra. Callers choose low, medium, or high effort and can steer a run toward NDCG, F1, or a set recall target. The API is available now, including a serverless option with zero data retention. SID also states SOC 2 and HIPAA coverage and offers self-hosted deployment in a customer VPC. SID was founded in 2023 by Maximilian-David Rumpf and Lotte Seifert and joined Y Combinator's Summer 2023 batch. Rumpf, the CEO, previously researched AI accelerator design at ETH Zürich. The lab, with offices in San Francisco and Zürich, lists Y Combinator, Canaan, Rebel, and General Catalyst as backers, and released its first retrieval model, SID-1, in December 2025. SID-2 extends that line with the finding that coordinated swarms improve retrieval quality per unit of compute more than fusing independent runs or increasing a single model's parameter count.
Tags
Questions
Frequently asked questions
The SID AI launch post reached 575K views on X, with 5.2K likes, 435 reposts, 286 comments, 4.3K bookmarks.
SID AI launched on X on November 17, 2026 with the message "Search model whose compute-scaling swarms return better results faster and cheaper".
SID AI has released SID-2, a search model for private data, and SID-2 Swarm, which runs that model as a coordinated team.
LaunchVideo.com rated the SID AI launch 3.80 out of 5: hook 3/5, editing / creativity 4/5, copy 4/5, sentiment of launch 4/5, distribution strategy 4/5.






