SID AI has launched SID-2, its second-generation agentic search mannequin, alongside SID-2 Swarm. SID-2 is an agentic search mannequin from SID AI that searches your information, reads outcomes and returns ranked doc IDs. It replaces fastened RAG pipelines with a mannequin that decides what to look subsequent.
TL;DR
- Dimension: Context window is 262k tokens per the report (256k on sid.ai/models).
- Runs on: Hosted API at platform.sid.ai, plus Baseten and SageMaker reference paths. SID’s latency figures assume 1 B200 GPU at 8 concurrent queries.
- Efficiency: A single agent matches Claude Opus 5.5 on retrieval high quality. Swarms beat each mannequin SID examined.
- Finest: SID-2 Swarm (12 brokers) scores 0.720 NDCG, above GPT-6 Astra at 0.704.
- Worst: SID-2 Low scores 0.556 NDCG, 0.148 beneath GPT-6 Astra. Astra nonetheless leads on California legislation, SEC filings and FiQA.
- Backside line: Finest: frontier-level retrieval at a fraction of the latency. Worst: no disclosed measurement, weights or unbiased evals but.
What’s SID-2 and the way does it work?
SID-2 is a small search specialist, not a common chat mannequin. A static pipeline runs rewrite, embed, search and rerank in a hard and fast order. SID-2 as an alternative searches, reads results, adapts and searches again.
Builders implement the search instruments towards their very own backend. A text_search device covers key phrase search and a search device covers vector search. The mannequin ends every episode by calling report_helpful_ids with a ranked record. No reindexing or data movement is required, since SID-2 solely sees what your instruments return.
SID-2 Swarm runs a number of brokers that message each other and fork their context into new brokers. Every agent explores a separate slice of the corpus. SID argues search positive aspects extra from studying extra paperwork than from reasoning more durable about each.
What are reasoning effort and Metric Steering?
SID-2 gives low, medium and excessive reasoning effort. Every stage was skilled as a separate professional, then merged with multi-teacher on-policy distillation. Metric Steering lets customers goal NDCG, F1 or Recall@ok, with ok from 1 to 50. On SID’s 506-question validation set, SID-2 at excessive effort scores 0.14 to 0.22 greater per metric than the bottom mannequin.
How was SID-2 skilled?
Coaching used 2-stage reinforcement studying. Stage 1 optimizes retrieval high quality solely. Stage 2 provides a multiplicative latency penalty that applies solely to constructive rewards. A proposer-solver self-play loop generated arduous queries towards a frozen SID-1 solver. SID stories a 6x training-throughput acquire over SLIME’s default setup.
How does SID-2 carry out on benchmarks?
SID constructed a 506-question benchmark throughout 19 corpora. These embrace HotpotQA, Enron emails, SEC filings, PubMed and BrowseComp-Plus. All numbers beneath are self-reported from Tables 2 to 4 of the report.
- SID-2 Medium: 0.585 NDCG in 3.11 s, versus Claude Opus 5.5 (medium) at 0.577 in 25.1 s.
- SID-2 Swarm, 6 brokers: 0.701 NDCG in 17.7 s, versus GPT-6 Astra (max) at 0.704 in 135 s.
- SID-2 Swarm, 12 brokers: 0.720 NDCG, the best rating within the desk.
- Baselines: BM25 scores 0.260 and vector search plus reranker scores 0.287.
Swarm leads on 11 of 19 corpora. On WixQA, a single high-effort agent matches Astra at about 232x decrease price. At its lowest effort, SID-2 beats 4 parallel SID-1 runs with 13x fewer FLOPs.
(operate(){var f=doc.getElementById(‘mtp-sid2-frame’);window.addEventListener(‘message’,operate(e){if(e.information&&e.information.kind===’mtp-sid2-h’&&e.supply===f.contentWindow){f.type.peak=e.information.h+’px’;}});})();
#mtp-sid2-wrap p:empty,#mtp-sid2-wrap hr,#mtp-sid2-wrap del,#mtp-sid2-wrap s{show:none!necessary;}
SID-2 vs GPT-6 Astra vs Claude Opus 5.5
| Characteristic | SID-2 (medium) | SID-2 Swarm (6 agents) | GPT-6 Astra (max) | Claude Opus 5.5 (medium) |
|---|---|---|---|---|
| Kind | Agentic search mannequin | Multi-agent search | Basic LLM | Basic LLM |
| Parameters | Not disclosed | Not disclosed | Not disclosed | Not disclosed |
| Context | 262k (report) / 256k (site) | Effectively infinite (per SID) | Not disclosed | 1M |
| NDCG, SID benchmark | 0.585 | 0.701 | 0.704 | 0.577 |
| Inference time per question | 3.11 s | 17.7 s | 135 s | 25.1 s |
| Price per 1,000 queries | $6.54 | $263 | $2,141 | $131 |
| API value (enter / output per 1M) | $0.25 / $1.00 | $0.50 / $2.00 | $10 / $50 | $4 / $20 |
NDCG, time and price rows are measured by SID by itself benchmark. Retrieval backend time is excluded.
How can builders deploy SID-2?
SID-2 is obtainable by SID’s API and on Baseten. SDKs exist for Python (sid-sdk), TypeScript (@sid-ai/sid-sdk) and Go (github.com/sidhq/sid-go). A SageMaker helper covers AWS deployments. The reference implementation makes use of the OpenAI consumer format.
SID-2 prices $0.25 per million input tokens and $1.00 per million output tokens. Swarm prices $0.50 and $2.00. Cached enter is 10% of the enter value.
Key Takeaways
- SID-2 is an RL-trained agentic search mannequin that returns ranked doc IDs.
- SID-2 Medium matches Opus 5.5 retrieval at 0.585 NDCG in 3.11 s.
- 12-agent SID-2 Swarm leads SID’s benchmark at 0.720 NDCG.
- Pricing begins at $0.25 per million enter tokens, with no reindexing.
- All outcomes are self-reported; measurement and weights are undisclosed.
Try the SID-2 technical report, the docs and the model page. You’ll be able to verify the deployment choices and specs right here by our GPULeaderboard.com chrome extension.
Additionally, be at liberty to comply with us on Twitter and don’t neglect to hitch our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
Have to associate with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and many others.? Connect with us
The put up SID AI Releases SID-2 Agentic Search Model With 0.720 NDCG Swarm Mode appeared first on MarkTechPost.

