--- type: "post" id: "llm-deployment-strategy-and-automation-economics-analysis-b1ac42" slug: "llm-deployment-strategy-and-automation-economics-analysis-b1ac42" title: "Llm deployment strategy and automation economics" description: "T-Search frames long-horizon retrieval as a memory-selection problem rather than simply a larger-context or stronger-model problem." retrieval_nugget: "T-Search frames long-horizon retrieval as a memory-selection problem rather than simply a larger-context or stronger-model problem." published_at: "2026-09-28" updated_at: "2026-09-28" record_date: "2026-09-28" date_kind: "published_at" topics: [] source_urls: ["https://habr.com/ru/companies/tbank/articles/1060262"] schema_version: "newruntime-agent-readable-v0.2" stable_id: "post:llm-deployment-strategy-and-automation-economics-analysis-b1ac42" status: "published" visuals: [] editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:ef5269bb9d01c8118c8d91506d413b2d6cd356e84cc932ee7361833cc46127a2","reviewed_at":"2026-09-28T10:03:54.507Z","source_evidence_count":1,"verified_claim_count":4,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":4,"implication_count":2,"watch_condition_count":2,"related_record_count":0} analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"T-Search frames long-horizon retrieval as a memory-selection problem rather than simply a larger-context or stronger-model problem.","observed_facts":[{"text":"The source describes T-Search as an open agent-retriever built on Qwen3.6-35B-A3B and trained entirely on synthetic data.","source_urls":["https://habr.com/ru/companies/tbank/articles/1060262"]},{"text":"The source says T-Search returns ranked evidence chunks with rationales rather than a final answer.","source_urls":["https://habr.com/ru/companies/tbank/articles/1060262"]},{"text":"The release includes TRuST, a manually assembled Russian benchmark for complex agentic search, and SynthComp in Russian and English.","source_urls":["https://habr.com/ru/companies/tbank/articles/1060262"]},{"text":"The harness uses roughly 32,000-token rounds and blocks search once a round reaches 75% context use.","source_urls":["https://habr.com/ru/companies/tbank/articles/1060262"]}],"mechanism":"The mechanism is explicit evidence preservation across bounded contexts, combined with constraints that discourage empty resets and premature completion.","why_now":"The immediate relevance is that teams building retrieval systems are confronting context noise as a systems-design issue, not only a model-selection issue.","implications":["For product teams, the implication is that retrieval architecture may need separate evaluation from answer-generation quality.","Decoupling retrieval from final-answer generation can allow product-specific generation policies without retraining the retriever."],"evidence_boundary":"The source is a vendor-authored technical account and does not independently establish comparative production performance, cost, or generalization beyond its reported setup.","watch_conditions":["Watch whether the released model, code, and benchmarks enable independent replications that compare this memory strategy against summarization, full resets, and larger-context baselines.","Watch whether benchmark outcomes hold on production corpora with changing indexes and real latency constraints."],"related_records":[],"new_branch_reason":"No prior New Runtime record with usable substantive detail was supplied; this release warrants a distinct record because it introduces a specific retriever, harness design, and accompanying benchmarks."} editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]} routes: {"html":"https://newruntime.com/posts/llm-deployment-strategy-and-automation-economics-analysis-b1ac42/","markdown":"https://newruntime.com/posts/llm-deployment-strategy-and-automation-economics-analysis-b1ac42.md","json":"https://newruntime.com/posts/llm-deployment-strategy-and-automation-economics-analysis-b1ac42.json"} --- # Llm deployment strategy and automation economics ## Retrieval answer T-Search frames long-horizon retrieval as a memory-selection problem rather than simply a larger-context or stronger-model problem. T-Search frames long-horizon retrieval as a memory-selection problem rather than simply a larger-context or stronger-model problem. The source describes T-Search as an open agent-retriever built on Qwen3.6-35B-A3B and trained entirely on synthetic data. It returns ranked evidence chunks with rationales, not a completed answer. The release also includes TRuST, described as a manually assembled Russian benchmark for complex agentic search, plus SynthComp in Russian and English. Its operating mechanism is a multi-round ReAct loop: the agent searches a corpus, explicitly saves useful chunks and coverage state, then begins the next round with selected memory instead of the full tool-call history. The source says each round has roughly a 32,000-token budget; once 75% is filled, search is blocked and the agent must either finalize or save evidence and advance. The harness also requires at least five searches and one saved chunk before an early round transition. The mechanism is explicit evidence preservation across bounded contexts, combined with constraints that discourage empty resets and premature completion. The immediate relevance is that teams building retrieval systems are confronting context noise as a systems-design issue, not only a model-selection issue. The source is a vendor-authored technical account and does not independently establish comparative production performance, cost, or generalization beyond its reported setup. For product teams, the implication is that retrieval architecture may need separate evaluation from answer-generation quality. Watch whether the released model, code, and benchmarks enable independent replications that compare this memory strategy against summarization, full resets, and larger-context baselines. Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json ## Continue in the graph Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`. - collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis. - collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph. - whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)