Back home

SID-2

x+web 0.94
Visit source

Intro

SID-2 is the new search model SID.ai introduced on 17 November 2026, built on compute-scaling search swarms. Founder Max Rumpf posted “Introducing SID-2” at 18:27 UTC, called it “a new search model built on compute-scaling swarms,” and said the swarms “find better results than Opus and Astra, while being 8x faster and 20x lower cost.” The same post argued that by 2030 AI will write an internet’s worth of text every 24 hours, that current search pipelines were built on hand-designed rules, and that “SID-2 is available on our API today.”

A follow-up post about half an hour later said SID-2 and SID-2 swarm are available at sid.ai, with a full technical report. The SID research library lists “SID-2 Technical Report: Scaling Search Swarms.” Platform docs also include a research-preview page at https://platform.sid.ai/docs/research-preview/sid-2.

Do not confuse it with SID-1. The same research page lists “Introducing SID-1” and “SID-1 Technical Report: Test-Time Compute for Retrieval” under December 2025. SID-2 is the version announced on 17 November, with the emphasis on scaling search swarms.

Advertisement

Explain

SID.ai positions SID-2 as a search and retrieval model, not a general chat model. The official comparison in the launch post is to Opus and Astra: Rumpf claims the swarms find better results while being 8× faster and 20× lower cost. Those figures come from the official tweet; evaluation settings, datasets, and exact opponent versions should be taken from the technical report. This entry does not treat the tweeted multiples as independently audited benchmarks.

The technical pitch is that search should scale as a swarm, because retrieval pipelines built on hand-designed rules will not keep up with the coming volume of AI-written text. The research page titles the report “Scaling Search Swarms.” Documentation describes an agentic retrieval model that searches a customer’s data and returns ranked IDs for the documents needed to answer a question. SID-2 swarm is named in the follow-up post as a form that launched alongside SID-2.

On availability, the first-party claim is that it is available on the API today, at sid.ai. Quotas, pricing, and research-preview limits should be read from the platform docs. When citing the comparison, note that the tweet does not spell out the full names or versions of Opus and Astra, so those should not be silently expanded into a specific Claude or GPT release.

Usage

According to the official announcement, SID-2 and SID-2 swarm are available at sid.ai. The introduction is Introducing SID-2, the technical report is listed at sid.ai/research, and preview documentation is at platform.sid.ai/docs/research-preview/sid-2.

The intended use is as a retrieval and ranking service: run agentic search over your own or workspace data and get ranked documents (or document IDs) needed to answer a question, rather than treating SID-2 as another general-purpose chatbot. Whether to use the swarm, and how to trade compute against latency, should follow the official API and docs.

If a source is about SID-1 or the December 2025 Test-Time Compute for Retrieval report, that is the previous generation, not this SID-2 release. When repeating the official claim of better results than Opus and Astra at lower latency and cost, go back to the settings in the technical report rather than citing only the tweeted multiples.