PapersSelected11月11日 01:44

Route Anything 研究：模型路由平均带来 8.82 倍单位成本质量收益

rohanpaul_ai@rohanpaul_ai

Post

View the post

Why we picked it

Berkeley 这篇把路由研究扩展到 25 个基准和 452 个模型，结论很实在：路由基本不提质，但省钱确实猛，编程场景能比固定调 Opus 4.6 省 25 倍。做 LLM 成本优化的工程师建议直接看。

UC Berkeley 等机构发布 Route Anything 研究，在 452 个候选模型和路由策略、25 个基准上跑出 374,401 次带评分的运行。结论是路由平均带来 8.82 倍的单位成本质量收益，质量损失低于 0.5 分。

Body

路由的机制是为每个输入挑选最合适（通常更便宜）的模型处理。研究将路由从纯语言模型扩展到机器人控制等任务。理论上完美的逐输入路由能比最佳单一模型高 5.4 分，但实际学习到的路由在 16 个基准上平均只提升 0.3 分，说明路由收益主要来自省成本而非提质。

具体到编程任务，路由达到 always 调用 Opus 4.6 的 25.5 倍美元效率，代价仅 0.44 分。论文建议把路由定位为“质量接近时切到便宜模型”的工具，并在同一批输入上对比候选后再信任路由收益。

Original · rohanpaul_ai

Great new study on model routing.

From University of California, Berkeley + other labs

Finds that model routers pay off significantly by saving money, averaging 8.82X more quality per unit cost while losing under 0.5 points.

Routers pick which model handles each input, hoping to match the best model for less or beat it. Most routing research has studied only language models.

Route Anything ran 452 candidate models and policies on the same inputs across 25 benchmarks, from text QA to robot control. They trained routers on these 374,401 scored runs and measured both quality and cost.

A perfect per-input router would beat the best single model by 5.4 points, but learned routers gained only 0.3 across 16 benchmarks. On coding tasks, routing reached 25.5× the dollar efficiency of always calling Opus 4.6 at a 0.44-point loss.

Use routers mainly to move work to cheaper models when quality is close, and compare candidates on the same inputs before trusting gains.

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Route Anything, 模型路由, Opus 4.6, UC Berkeley.

Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).

#Route Anything #模型路由 #Opus 4.6 #UC Berkeley #推理成本

Keep reading

Recent stories with the same tag or section

Papers BuildersSelected01:19

UC Berkeley's Agency Cuts Agent Harness Costs by Up to 35%

The practical shift here is that cost optimisation moves from model selection to infrastructure — a middleware layer can be reused across Claude Code, Codex and other harnesses. Engineering leads running production agents should audit trace logs for repeated context reads and unnecessary tool calls before upgrading to a costlier model. Teams in cost-sensitive markets, including ASEAN startups paying USD-denominated API bills, gain a lever that compounds across every harness they deploy.

#Agency #UC Berkeley #agents #coding assistants

DAIR.AI@dair_aiSource

Papers BuildersSelected01:18

InfraRSI Lifts LLM Training MFU from 17.4% to 30.0%

The non-obvious takeaway is that the hard part of agent-driven infra optimisation is verification, not code generation — the contract-and-recheck design is what makes the 2.75x speedups real. ML engineers and platform teams running multi-GPU training can apply the same contract-based evaluation to their own agent workflows. The cross-GPU memory of learned models suggests reusable optimisation knowledge across heterogeneous clusters.

#InfraRSI #training infrastructure #agents #GPU optimisation

DAIR.AI@dair_aiSource

Papers01:17

Chinese contentAI Futures Project 更新时间线：超级人类 AI 研究员或于 2027 年出现

AI 2027 那份预测报告的作者们把时间线提前了 5 个月，认为 2027 年就会出现能自己做 AI 研究的超级系统。不管你信不信这个预测，这份报告此前引发了大量讨论，看看他们为什么改主意挺有意思。

#AI Futures Project #AI 2027 #Superhuman AI Researcher #AI预测

Event

Andrew Curran@AndrewCurran_Source

Papers Policy00:56

Chinese contentStanford AI100 发布第三份报告《AI Goes Mainstream》，回顾 2021 年以来 AI 变化

Stanford AI100 这个百年研究计划刚出了第三份报告，专门梳理 2021 年以来生成式 AI 怎么一步步走进科研和日常生活。跟前两份比，这版把生成式 AI 的规模化当成主线来写，做政策或研究的人可以直接拿来当参考。

#Stanford HAI #AI100 #生成式AI

Stanford HAI@StanfordHAISource

Papers BuildersMulti-sourceSelected00:52

Nvidia paper identifies intrusion heads behind RAG distraction

The non-obvious finding is that distractor damage is localised in a handful of attention heads, so a lightweight patch can match most of what a costly retraining effort would achieve. Teams running RAG or agent systems on open-weight models can implement the mask as an inference-time intervention without touching training pipelines. Compared with context-cleaning heuristics that filter noisy documents before inference, head masking operates inside the model and needs only a one-time identification of the offending heads per model. Practitioners in ASEAN serving models under cost constraints benefit because the fix adds no training expense.

#Nvidia #RAG #attention heads #open-source models

Event