Skip to content
OpenSmartRoute

Research Foundations

Problem statement, literature by family and which module borrows which idea.

docs/RESEARCH.md

This document is the evidence base for OpenSmartRoute. It (1) states the routing problem formally, (2) surveys the literature by family, (3) compares the systems that exist today, (4) lists the gaps that motivate this project, and (5) maps every borrowed idea to the module that implements it. It is maintained as the design changes; see ROADMAP.md for what is still open.


1. Problem statement#

A request $x$ arrives with context $c$ (tenant, session, constraints). A catalogue $\mathcal{T}$ of targets is available — models, agents, skills, personas, tools, workflows, humans. Each target $t$ has a cost $c_t(x)$, latency $\ell_t(x)$ and an unknown quality $q_t(x)\in[0,1]$. Routing chooses

$$ t^* = \arg\max_{t\in\mathcal{A}(x)} ; \mathbb{E}\big[q_t(x)\big] - \lambda_c, c_t(x) - \lambda_\ell, \ell_t(x) $$

where $\mathcal{A}(x)\subseteq\mathcal{T}$ is the admissible set after hard constraints (data boundary, region, PII, budget, SLO). Three things make this hard:

  1. $q_t(x)$ is never observed before routing — it must be predicted from $x$ and from $t$'s history.
  2. $\mathcal{T}$ changes — targets are added, deprecated, and degrade; the router must be inductive.
  3. Feedback is delayed, partial and noisy — a thumbs-up, a retry, a handoff, or nothing.

The survey by Varangot-Reille et al. (2025) formalises this as a performance–cost optimisation and classifies solutions as pre-generation (predict, then call one target) or post-generation (call, verify, escalate). OpenSmartRoute does both: strategies are pre-generation; Cascade and the planned multi-round executor are post-generation.


2. Literature by family#

2.1 Supervised routers (predict quality from the prompt)#

WorkCore ideaResultLimitation
FORC — Šakota, Peyrard & West, WSDM 2024A meta-model predicts each LM's performance on an input; assign under a cost budget or minimise cost at a target qualitymatches largest LM at −63 % cost on 14 datasetsfixed LM pool; retrain when the pool changes
Hybrid LLM — Ding et al., ICLR 2024Predict query difficulty; route to small vs large with a tunable quality threshold−40 % large-model calls at equal qualitybinary pool
RouteLLM — Ong et al., 2024Train routers on human preference (Chatbot Arena) with data augmentation; four router types (BT, matrix factorisation, BERT, causal LLM)>2× cost reduction; routers transfer to new strong/weak pairspairwise (two targets)
Zooter — Lu et al., 2023Distil reward-model scores on training queries into a routing function; tag-based label smoothingbeats best single model on 26 subsetsneeds an offline reward model
RouterDC — Chen et al., NeurIPS 2024Encoder + learnable LLM embeddings trained with dual contrastive losses (sample–LLM, sample–sample)+2.8 % ID / +1.9 % OOD over best single LLMhandles "several are fine" better than pointwise
Arch-Router — Tran et al., 20251.5 B model maps query → user-defined domain / action; routing preferences encoded declaratively; new models added without retrainingSOTA on preference alignment for conversational datarouting policy still hand-authored
Prompt-to-Leaderboard — Frick et al., 2025LLM outputs Bradley–Terry coefficients per prompt → prompt-specific leaderboard; optimal router follows#1 on Chatbot Arena (Jan 2025); power-law scalingneeds preference data at scale

2.2 Inductive / representation-based routers (generalise to unseen targets)#

WorkCore ideaWhat we took
GraphRouter — Feng, Shen & You, ICLR 2025Heterogeneous graph (task, query, LLM); edge prediction of effect + cost; new LLM = new nodetargets are described by declared capabilities, not learned IDs
UniRoute — Jitkrittum et al., 2025Represent an LLM by its prediction vector on a representative prompt set; cluster-based routing; excess-risk bound; 30+ unseen LLMsexamples on a target; nearest-example similarity
EmbedLLM — Zhuang et al., 2024Encoder–decoder learns compact LLM embeddings that forecast benchmark performanceroadmap v0.5: target embeddings from outcomes
IRT-Router — Song et al., ACL 2025Item Response Theory: target ability θ, query difficulty b, discrimination a; interpretable; semantic warm-up for cold startmath.irt, learning.IRTStrategy

2.3 Online / bandit routers (learn from live feedback)#

WorkCore ideaWhat we took
MixLLM — Wang et al., NAACL 2025Contextual bandit; tag-enhanced query embedding; per-LLM quality & cost predictors; meta-decision for quality/cost/latency; continual training; handles pool changesLinUCBStrategy on the signal vector; Objective weights
C2MAB-V — Dai et al., 2024Combinatorial MAB with versatile reward models and cost constraints; regret & violation boundsCostAwareBandit (Lagrangian budget)
Thompson sampling / UCB1 (Thompson 1933; Auer et al. 2002)Posterior sampling / optimism for explorationThompsonBeta, UCB1
LinUCB (Li et al., WWW 2010)Linear payoff with confidence ellipsoidLinUCB

2.4 Cascades and multi-round routers (post-generation)#

WorkCore ideaWhat we took
FrugalGPT — Chen, Zaharia & Zou, 2023Call cheap first; a scorer decides whether to escalateCascade with pluggable quality_gate
AutoMix — Aggarwal et al., NeurIPS 2024Few-shot self-verification + POMDP router over verification noise; −50 % costroadmap v0.6: POMDP cascade
Router-R1 — Zhang, Feng & You, NeurIPS 2025Router is itself an LLM; interleaves think / route actions across rounds; RL with format + outcome + cost reward; conditions on model descriptors (price, latency, examples)LLMJudgeStrategy (conditioned on the same descriptors), confidence-gated escalation; roadmap v0.6 multi-round
LLM-Blender — Jiang, Ren & Lin, ACL 2023PairRanker over candidate outputs + GenFuseraggregation step for the multi-round executor

2.5 Multi-agent, tool and skill routing#

WorkCore ideaWhat we took
MasRouter — Yue et al., 2025Multi-Agent System Routing: choose collaboration mode → agent roles → LLM per role via cascaded controllers; −52 % overheadRoutePlan slots (persona → skill → model) filled by sub-routing
ToolLLM — Qin et al., 202316 k APIs; neural API retriever recommends tools per instruction; DFS decision treetools are targets with examples; retrieval = SimilarityStrategy
AnyTool — Du et al., 2024Hierarchical API retriever (category → tool → endpoint) + self-reflection; revised evaluation protocolroadmap v0.7: MCP catalogue import with hierarchical narrowing
Agent Skills (Anthropic, 2025) / skills as first-class unitsSkills are declared with a description and triggersTargetKind.SKILL with capabilities.actions

2.6 Sequential decision making for conversations#

Dialogue policy as an MDP dates to Levin, Pieraccini & Eckert (2000) and is the basis for POMDP dialogue systems (Young et al., 2013). OpenSmartRoute applies the same lens to who answers next: MarkovChain learns intent transitions, RoutingMDP runs value iteration with reward = quality − λ·cost, and MarkovStrategy scores targets by their value on the predicted next state, not only the current one.

2.7 Capacity and queueing#

Skills-based routing in contact centres is a mature field (Koole & Mandelbaum 2002). Its tools carry over directly to pools of humans or long-running agents: Erlang-C for P(wait) and mean wait in M/M/c, Little's law for in-flight sizing, and Kingman's G/G/1 approximation for bursty single endpoints. Implemented in math.decision.

2.8 Adversarial robustness of routers#

Rerouting LLM Routers (Shafran, Schuster, Ristenpart & Shmatikov, 2025) defines LLM control-plane integrity and shows query-independent confounder gadgets — short token sequences that push any query to the strong model in white- and black-box settings, with low perplexity (so perplexity filtering fails). Consequences for us: routing signals are computed on the cleaned head of a prompt (security.InputGuard), the guard reports the suspicion in the trace, and the roadmap keeps a learned detector open. See SECURITY.md.

The 2026 literature widens the control-plane threat model beyond gadgets:

WorkThreatConsequence for us
ROPE — Aug 2026Indirect prompt injection through tool results steering state-changing tool calls; deterministic origin check on sensitive parameters keeps attack success at 1.6–2.6 % with 82–100 % utilityorigin policy for tool parameters inside executed RoutePlans (v0.9)
Verifiable manifest signing for MCP pipelines — Jan 2026Tampered or drifting tool manifestssigned-manifest verification at import
"Recognition Without Enforcement" — Aug 2026Models recognise unsafe tool requests but still execute themenforcement lives in policy, not in the model
"Beyond Max Tokens" — Jan 2026Resource amplification through tool chainsper-task depth / token caps
"When Safety Routing Breaks" — Sep 2026Routing itself as the safety bypassadversarial routing regression suite

2.9 Benchmarks and evaluation methodology#

BenchmarkScaleUse here
RouterBench — Hu et al., 2024405 k inference outcomes, 11 LLMs, 8 datasets; defines the quality–cost frontier and AIQeval.cost_quality_frontier, area_under_frontier; eval.collect lists the source (routerbench, gated: the Hub serves no data files - load a local copy with DatasetCollector.store)
RouterEval — Huang et al., 2025200 M records, 8 500 LLMs, 12 evaluations; shows model-level scaling — a good router improves as the pool grows and can beat the best single modelmotivates large catalogues; eval.collect source routereval (gated: needs a Hub token)
LLMRouterBench — Jan 2026400 k instances, 21 datasets, 33 models, 10 baselines; many routers, including commercial ones, do not reliably beat simple baselines; large oracle gap from model-recall failures; larger ensembles show diminishing returns vs curation; latency-aware analysisroadmap v0.4: mandatory baselines, curation report
LLMRouter / xRouteBench — Feng & You, Aug 2026Routing as a sequential decision with five components (context encoder, model encoder, scoring, decision rule, learning signal); 16+ routers; single/multi-turn, agentic and personalised tasks; learned routers +14.6 % over the best fixed modelcomponent ablation report; personalised split
RouterXBench / ProbeDirichlet — Feb 2026Three axes: router ability, scenario alignment, cross-domain robustness; hidden-state Dirichlet router +16.7 / +18.9 %cross-domain split; math.DirichletProbe + strategies.HiddenStateStrategy
RouterArena — 2026Public leaderboard with arena score, accuracy and cost per 1 k queriesv1.0 submission target
"How Much of the Routing Gap Is Real?" — Jul 202612–36 % of the router-to-oracle gap on open pools is single-draw label noise; multi-sample oracle ($k\ge 20$) and best-of-K recover the floor--oracle-samples, reproducible-headroom figure
"Most of the LLM Routing Gap Is Task Type" — Aug 202621 methods within a fraction of a point; a static per-task-type table answered 262/294 at $3.33 vs 245 at $7.69 for the best single model; 5.4 % run-to-run flipsTaskTableStrategy; --repeats
"When is Routing Meaningful?" — Jul 2026Hierarchic social entropy for pool diversity; <10 behaviourally distinct agents suffice; kNN routers collapse under paraphrase, prompted routers stay stablediversity / coreset report; paraphrase-robustness score
MixInstruct (LLM-Blender)oracle pairwise comparisonspreference data format for FeedbackStore.export_preferences
SkillFeed-Bench, Task2MCP, ComplexMCP — 2026counterfactual profile-conditioned skill routing; MCP server recommendation; multi-server tool tasksv0.7 discovery evaluation

2.10 Risk control, calibration and abstention#

WorkCore ideaWhat we take
RACER — Feb 2026α-VOR: output a set of models with distribution-free mis-routing risk ≤ α, plus abstention; nested sets from concentration boundsRouteDecision.candidate_set, abstain → HUMAN
RouteNLP — ACL 2026 (industry)Conformal prediction to set cascade thresholds; distillation–routing co-optimisation by clustering escalation failures; −58 % cost, p99 1847 → 387 msconformal escalate_below; escalation-failure clustering feeds osr train
CR² — May 2026Cost-aware, risk-controlled routingsame family
Temperature scaling (Guo et al. 2017), isotonic regression, ECEPost-hoc probability calibrationfitted confidence instead of fixed $\tau$
Semantic entropy (Kuhn et al. 2023; Farquhar et al. 2024), P(True) (Kadavath et al. 2022)Response-side uncertaintysignals.semantic_entropy(), response_uncertainty(), UncertaintyGate, EventTrigger
Learning to defer (Madras et al. 2018; Mozannar & Sontag 2020; Verma et al. 2023)Classifier and expert trained jointly with a consistent surrogate; the expert has its own cost and accuracyTargetKind.HUMAN as a deferral option

2.11 Multi-turn, task-level and step-level routing for agents#

WorkCore ideaWhat we take
TRACE-Router — Jul 2026Agentic tasks have one delayed, task-level outcome; assign a backend once at admission with a contextual bandit, pin it, learn from the terminal reward (accuracy + latency); +7–8 pts on τ²-Bench, +7.1 pts / −36 % latency on Terminal-BenchOutcome.task_id, admission-time routing, delayed-feedback bandits
MTRouter — ACL 2026History–model joint embeddings; turn-level utility from logged trajectories; beats GPT-5 on ScienceWorld at −58.7 % cost with fewer switcheslearning.HistoryTargetStrategy: logistic model over h * e_t (recency-weighted hashed history, catalogue target embedding) with shared weights and per-target bias; incumbent bonus / failed-turn penalty from context["last_target"]
ProgRouter — EMNLP 2026 FindingsMulti-view task-progress scorer + dual-path progress predictor + meta-gating; step-wise routing under time & cost budgetsprogress-guided step routing
RLCascadeRouter — Aug 2026Cascade routing as an MDP with stop / select actions trained on trajectory returns; quality-estimator-free; prediction loss ≠ decision loss; unseen models without retrainingcascade rewrite on RoutingMDP
Bayesian self-escalation — Aug 2026Intra-generation delegation as Bayesian optimal stopping over a learned competence posterior; time-varying threshold; $1/\sqrt{n}$ regret in calibration trajectoriesstrategies.SelfEscalation + wrap_stream(): Beta competence posterior per streamed chunk, stop when loss(stop) < loss(continue)
TACIT-Switch — Aug 2026Permanent handoff learned from interval-censored teacher intervention times; mixture-cure threshold model; +7.4–11.1 pplearning.MixtureCureModel (Weibull mixture-cure, censored MLE) + learning.HandoffPolicy
JiSi — Jan 2026Query–response mixed routing, support-set aggregator selection, adaptive routing ↔ aggregation switch; 10 open LLMs beat Gemini-3-Pro at 47 % costsignals.DraftResponseSignal (query–response features); strategies.MixtureOfAgents (routing / aggregation switch)
GMTRouter — EMNLP 2026 FindingsHeterogeneous graph (user, LLM, query, response, turn); few-shot personalisationlearning.UserAdaptiveStrategy: user → neighbour users → global shrinkage over hashed profile vectors
TRouter — ACL 2026Task-taxonomy-guided synthetic data for cold start; latent task-type variables with prior regularisationosr train --synth
SCX Router — Sep 20260.6 B decoder-KV classifier predicts task type, difficulty, reasoning mode and output length before generation; task ontology (23 families / 115 types / 345 subtypes / 30 domains); separates learned prediction from per-task policies (eligibility, cost, cache, safety, sovereignty)ontology-based task classifier; policy/prediction separation matches ours
Think When Needed — SIGIR 2026Pre-generation router head decides Think vs Non-Think; operating point chosen on the validation Pareto frontier; +6.3 % NDCG@10 at −49.5 % tokenseffort as a routable dimension
Agent-as-a-Router, SWE-Router, "Routing least learnable where most valuable" — 2026Routing gains shrink on the hardest agentic taskseval.headroom.learnability_by_difficulty: per-difficulty-bucket headroom and noise floor
BudgetMem, Gated-Memory Routing — 2026Routing applied to agent memory tiersstrategies.MemoryRouter (tiers with capacity / cost / latency, learned ImportanceGate, budgeted recall)

2.12 Capability discovery at scale (models, agents, tools, skills)#

WorkCore ideaWhat we take
Enrich-Retrieve-Rank — Aug 2026MATS (models · agents · tools · skills) discovery; offline enrichment + retrieve-then-rank; in-context routing Match@1 falls 0.85 → 0.12 at N = 7 278 while retrieve-then-rank holds 0.81 → 0.39; crossover ≈ N = 500; 70× cheaper than full contexttwo-stage narrowing switched on by catalogue size
SCOUT — Aug 2026 (PayPal)MCP gateway with tool_search / execute_tool meta-tools; BM25 + dense with RRF; 140 k → 1.3 k tool tokens (99 %) across 2 000+ tools / 200+ serversmeta-tools; hybrid retrieval
Task2MCP / T2MRec — 2026MCP server recommendation as retrieval + rank under engineering constraintsadapters.recommend_servers() over ServerCards with constraint filters
SchemaRouter, CacheRouter, Nexus, ToolDNS, Semantic Tool Discovery — 2026Field-aware matching, cache-preserving tool subsets, retrieval-decoupled routing, DNS-based discoverydiscovery.schema_match() / SchemaAwareStrategy; discovery.CachePreservingSelector; DNS discovery not adopted
Best Prefix Selection — Aug 2026Skill selection as submodular benefit minus context penalty under a token budget; $(1-1/e,,1)$ bicriteria greedy; 0.73 vs 0.20–0.52 success, −28 % tokensRoutePlan.skills set selection
SkillFeed — Aug 2026Profile-conditioned skill retrieval with a counterfactual benchmark; +35 pt when the profile matterslearning.SkillAffinityselect_skill_set(relevance=); profile-swap evaluation
Compositional skill routing, CaSKG, systems foundation for agentic skills — 2026Skill graphs and compositiondiscovery.SkillGraph (requires / conflicts / composes + learned co-usage; compose())

2.13 Non-stationarity, budgets and latency#

WorkCore ideaWhat we take
Drift-Aware Sparse Routing — Sep 2026Non-stationary sparse contextual routing with multiple knapsack constraints; shadow-audit stream for pruned arms; pessimistic reward / optimistic cost; online shadow prices; regret $\tilde O(T\sqrt{s/\rho W} + W V_T + \sqrt T)$multi-knapsack CostAwareBandit
OrcaRouter — May 2026LinUCB over lexical + sentence embeddings; offline full-information reward matrix → ridge per arm → online bandit updates; #2 on RouterArenaoffline warm start for LinUCBStrategy
Latency-aware routing — Jul 2026TTFT estimator that simulates batching; joint latency / accuracy / cost; +40 % utility at equal latencyqueue-aware latency in HealthStrategy
HW-Router — DAC 2026; energy characterisation — IISWC 2026Hardware- and energy-aware routingmath.EnergyModel (per-target Wh regression, gCO2) and hardware_profile() priors; energy as a knapsack resource
Federate the Router — Jan 2026Federated MLP / K-means routers from decentralised sparse evaluationsmergeable learner state
D-UCB / SW-UCB (Garivier & Moulines 2011), delayed-feedback bandits (Joulani et al. 2013), bandits with knapsacks (Badanidiyuru et al. 2013), dueling bandits (Yue et al. 2012), multi-objective bandits (Drugan & Nowé 2013)Classical machinery for drift, delay, budgets, preferences, Paretoforgetting, delayed credit, knapsacks, pareto_front
Off-policy evaluation: IPS, SNIPS, doubly robust (Dudík et al. 2011)Score a new policy on logged decisionseval OPE; propensity logging

2.14 Economics and mechanism design#

WorkCore ideaWhat we take
EA-RAM — Aug 2026Reverse auction: providers bid self-predicted success probability + cost; Bayesian incentive compatible under dual error; explicit welfare-loss boundstrategies.AuctionStrategy: bias-corrected claims, second-price payment
Strategy auctions for small agents — ICML 2026; coalition formation and communication pricing in skill-based agentic AI — 2026Market mechanisms among agentsresearch track
Dominant-resource fairness (Ghodsi et al. 2011)Fair multi-resource allocationtenant fairness under shared budgets

3. Systems landscape (open source, September 2026)#

SystemTargetsLearningExplainable traceHard policyDeps
RouteLLM (lmsys)2 LLMsoffline (preference)nonotorch
vLLM Semantic RouterLLMs (MoM)classifier (offline)partialyes (PII, jailbreak, location)Go/Rust/Python, models
Aurelio semantic-routerroutes (intents)embeddingsnonoembeddings
NVIDIA llm-routerLLMsclassifiernonoTriton
LiteLLM / Portkey / aisix / nyroLLMs (gateway)none (rules, fallback)norate/budgetservice
NadirClaw / Misceo / ClawRouter / nexus-llm-routerLLMs (cheap↔premium)heuristicsnobudgetPython/TS
LLMRouter (ulab-uiuc)LLMs16+ trainable routers, xRouteBenchnonotorch
SCX Router (checkpoint)LLMs0.6 B zero-shot classifier + task ontologytask labelsper-task policiestorch
OrcaRouterLLMsLinUCB, offline warm startnonoPython, embeddings
SCOUT (MCP gateway)toolshybrid retrievalnogateway ACLsservice
OpenSmartRouteLLM · agent · skill · persona · tool · workflow · humanonline (bandit, IRT, BT, LinUCB, MDP) + optional judgeyes, per strategyyes, before scoring0 in core

Gateways and OpenSmartRoute are complementary: the gateway translates protocols and executes; the router decides.


4. Gaps this project addresses#

  1. Heterogeneous targets. All surveyed routers choose among LLMs. Real systems choose among a model, an agent with tools, a deterministic skill, a persona layered on a model, or a human. Treating these uniformly (one RouteTarget contract, one policy layer, one learning loop) is the central design decision.
  2. Composition. MasRouter shows that mode + role + model beats picking a model. RoutePlan does this with declarative slots and no training.
  3. Inductive by default. GraphRouter/UniRoute/IRT show it is possible; most shipped routers still retrain on pool change. Every strategy here works from declared capabilities on day one and learns from day two.
  4. Hard constraints before soft scores. Cost/quality routers routinely trade away privacy or region. Policy runs first and cannot be outweighed.
  5. Control-plane integrity. Almost no router considers adversarial inputs; Shafran et al. show they must.
  6. Explainability. Regulated deployments need to say why a request went where it did. Every decision carries per-strategy scores and rationales.
  7. Honest evaluation. Benchmarks in 2026 show most routers do not beat a static task table and that a third of the "oracle gap" is label noise. The evaluation harness must make these baselines and noise floors unavoidable, or learned strategies will be adopted on illusory gains.
  8. Task-level supervision. Agentic workloads reward the task, not the call. The data model must carry task_id and trajectory position so learners can be credited correctly.
  9. Discovery at scale. Flat similarity over thousands of tools and skills collapses; two-stage retrieve-then-rank and set-valued skill selection under a token budget are required.
  10. Risk, not just cost. Set-valued decisions with abstention and conformal thresholds give operators a guarantee to sign off on; a tuned scalar threshold does not.

5. Idea → implementation map#

IdeaSourceModule
Difficulty-based routing with quality floorHybrid LLMsignals.ComplexitySignal, Objective.quality_floor
Domain/action preferences, pinned rulesArch-Routersignals.DomainActionSignal, strategies.RulesStrategy
Example-based target representationUniRoute, GraphRouterRouteTarget.examples, strategies.SimilarityStrategy
Ability × difficulty modelIRT-Routermath.irt, learning.IRTStrategy
Per-domain pairwise preferenceRouteLLM, P2Lmath.preference, learning.PreferenceStrategy
Contextual bandit on signal vectorMixLLM, LinUCBmath.bandits.LinUCB, learning.LinUCBStrategy
Budget-constrained explorationC2MAB-Vmath.bandits.CostAwareBandit
Cheap-first cascadeFrugalGPTstrategies.Cascade
LLM-as-router conditioned on descriptors, cost-awareRouter-R1strategies.LLMJudgeStrategy + escalate_llm_judge_below
Mode/role/model compositionMasRouterRouter._build_plan, RoutePlan
Conversation lookaheaddialogue MDPsmath.markov, learning.MarkovStrategy
Quality–cost frontier evaluationRouterBencheval
Confounder-gadget defenceRerouting LLM Routerssecurity.InputGuard
Capacity sizing for human poolsqueueing theorymath.decision.erlang_c, servers_for_sla
Drift detection on target qualityPage–Hinkley, ADWINmath.estimators, learning.AutoLearner
Static task-type table as default learner and baseline"Routing gap is task type"strategies.TaskTableStrategy (v0.4)
Multi-sample oracle, run-to-run flip rate"How much of the routing gap is real?"eval --oracle-samples, --repeats (v0.4)
Diversity / coreset and paraphrase robustness"When is routing meaningful?"eval reports (v0.4)
Ontology-based task / difficulty / reasoning-mode / length predictionSCX Router, TRoutersignals learned classifiers, osr train --synth (v0.4)
Calibrated confidencetemperature / isotonic scalingrouter confidence (v0.4)
Offline reward matrix → LinUCB warm startOrcaRouterlearning.LinUCBStrategy (v0.5)
Effort as a target dimensionThink When NeededRouteTarget.effort (v0.5)
Profile-conditioned routingGMTRouter, SkillFeedRouteRequest.profile (v0.5)
Task-level admission routing with delayed terminal rewardTRACE-RouterOutcome.task_id, delayed-feedback bandits (v0.6)
Cascade as MDP with a stop actionRLCascadeRouter, unified cascade routingstrategies.Cascade on math.markov.RoutingMDP (v0.6)
Progress-guided step routingProgRouterv0.6
Retrieve-then-rank narrowing, tool_search meta-toolEnrich-Retrieve-Rank, SCOUTadapters.mcp, two-stage selection (v0.7)
Submodular skill-set selectionBest Prefix SelectionRouter._build_plan (v0.7)
Conformal thresholds, set-valued routing with abstentionRouteNLP, RACERRouteDecision.candidate_set (v0.8)
Learning to deferMadras; Mozannar & SontagHUMAN deferral objective (v0.8)
Multi-knapsack budgets with shadow auditDrift-Aware Sparse Routingmath.bandits.CostAwareBandit (v0.8)
Forgetting in learnersD-UCB / SW-UCBmath.irt, math.preference, math.bandits (v0.8)
Queue-aware latency (TTFT)Latency-aware routingrealtime.HealthStrategy + math.decision (v0.8)
Off-policy evaluation, shadow modeIPS / DReval, EnterpriseRouter(shadow=) (v0.8)
Origin policy on tool parametersROPEsecurity (v0.9)
Contrastive (query, target) training over acceptable setsRouterDClearning.ContrastiveRouter, ContrastiveStrategy, acceptable_set
Reward-model-distilled soft labelsZooterlearning.soft_labels, ContrastiveRouter.fit(objective="distilled")
A small routing model distilled from the full ensemble, calibrated, learning onlineRouteLLM (matrix-factorisation router), Zooter, RouterDClearning.RouterSLM, distill_router, SLMStrategy
Self-improvement from live outcomes and public routing datasets without regressionsRouterBench, RouterEval; champion/challenger evaluationlearning.SelfImprover, eval.DatasetCollector, eval.rows_from_feedback
Live cost/quality priors and discovery of new modelsLLMRouterBench cost-quality frontieradapters.ModelCatalogue, adapters.WebKnowledge
Decision loss vs prediction loss; end-to-end policy gradientRouter-R1, RLCascadeRouterlearning.PolicyGradientStrategy, decision_reward, decision_regret
Oracle gap decomposition, target diversity, minimal catalogue, scaling curve"How much of the routing gap is real?", "When is routing meaningful?", RouterEvaleval.headroom (routing_headroom, target_diversity, min_catalogue, scaling_curve)
Learnability on the hardest tasksSWE-Router, Agent-as-a-Routereval.headroom.learnability_by_difficulty
Memory-tier routing under a token budgetBudgetMem, Gated-Memory Routingstrategies.MemoryRouter, MemoryTier, ImportanceGate
Modality-aware scoring and text-first escalationLatentRouter, modality escalation, CUA routingstrategies.ModalityStrategy, ModalityEscalation, request_modalities
Semantic cache as the cheapest targetGPTCache, vLLM semantic routerstrategies.SemanticCache, SemanticCacheStrategy
Annotator skill estimation and quorum selectionQUORUM; Dawid & Skene 1979strategies.AnnotatorPool, quorum_accuracy, HumanRoutingStrategy
Speculative draft/strong cascades with cancellationspeculative cascades; Differential Reasoning Routerstrategies.SpeculativeCascade, expected_mode_costs

6. Open research questions#

Every item of the roadmap research track now has a module (see the research-track table in ROADMAP.md); what remains open is measurement, not implementation:

  • How much routing headroom is real on a given catalogue once label noise and task-type effects are removed — eval.headroom.routing_headroom reports it; the open question is what the numbers say on your pool, and therefore when a learned router is worth its complexity over a static table.
  • How few, how different targets must be before routing beats the best single model (target_diversity, min_catalogue), and whether accuracy keeps scaling with catalogue size in heterogeneous pools (scaling_curve).
  • Prediction vs decision: decision_regret puts end-to-end RL (PolicyGradientStrategy) next to predict-then-optimise ensembles on the same rows; when does decision loss matter in practice?
  • Where to escalate: before generation (router), during (SelfEscalation), after (Cascade, MixtureOfAgents) or speculatively in parallel (SpeculativeCascade) — cost/quality frontiers of the four regimes on the same pool.
  • Contrastive vs pointwise training when several targets are acceptable (ContrastiveRouter vs TaskTableStrategy / LinUCBStrategy), and how much reward-distilled soft labels help over one-hot.
  • Routing for the hardest agentic work (web, SWE): learnability_by_difficulty on real agent traces.
  • Strategic providers: how AuctionStrategy's bias correction behaves against adaptive bidders; coalition pricing in multi-vendor catalogues.
  • Human pools: Dawid-Skene assumes conditionally independent annotators; correlated errors inside a quorum (AnnotatorPool.select_quorum) are not yet modelled.
  • Fairness under shared budgets; energy priors from hardware_profile() against metered fleets.

7. References#

Entries without author lists are 2026 preprints cited by arXiv identifier; the italic text is a descriptive title and the identifier is authoritative.

Surveys

  • Varangot-Reille C., Bouvard C., Gourru A., Ciancone M., Schaeffer M., Jacquenet F. Doing More with Less: A Survey on Routing Strategies for Resource Optimisation in LLM-Based Systems. JAIR 2025. arXiv:2502.00409

Supervised & preference routers

  • Šakota M., Peyrard M., West R. Fly-Swat or Cannon? Cost-Effective Language Model Choice via Meta-Modeling (FORC). WSDM 2024. arXiv:2308.06077
  • Ding D. et al. Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing. ICLR 2024. arXiv:2404.14618
  • Ong I. et al. RouteLLM: Learning to Route LLMs with Preference Data. 2024. arXiv:2406.18665
  • Lu K. et al. Routing to the Expert: Efficient Reward-guided Ensemble of LLMs (Zooter). 2023. arXiv:2311.08692
  • Chen S., Jiang W., Lin B., Kwok J.T., Zhang Y. RouterDC: Query-Based Router by Dual Contrastive Learning. NeurIPS 2024. arXiv:2409.19886
  • Tran C., Paracha S., Hafeez A., Chen S. Arch-Router: Aligning LLM Routing with Human Preferences. 2025. arXiv:2506.16655
  • Frick E. et al. Prompt-to-Leaderboard. 2025. arXiv:2502.14855
  • Stepanov I. et al. SCX Router: Streaming Zero-Shot Model Selection with a Decoder-KV Classifier and a Real-World Task Ontology. 2026. arXiv:2609.02292
  • TRouter: Task-Taxonomy-Guided Synthetic Data for Cold-Start LLM Routing. ACL 2026. arXiv:2604.09377
  • VDAR-Router: Verbalized Difficulty Retrieval for LLM Routing. 2026. arXiv:2607.18098
  • Scalable Prompt Routing via Latent Task Discovery. 2026. arXiv:2603.19415
  • Guo H. et al. Think When Needed: Model-Aware Reasoning Routing for LLM-based Ranking. SIGIR 2026. arXiv:2601.18146
  • When to Think Deeply. 2026. arXiv:2606.06745

Inductive / representation

  • Feng T., Shen Y., You J. GraphRouter: A Graph-based Router for LLM Selections. ICLR 2025. arXiv:2410.03834
  • Jitkrittum W. et al. Universal Model Routing for Efficient LLM Inference (UniRoute). 2025. arXiv:2502.08773
  • Zhuang R. et al. EmbedLLM: Learning Compact Representations of Large Language Models. 2024. arXiv:2410.02223
  • Song W. et al. IRT-Router: Effective and Interpretable Multi-LLM Routing via Item Response Theory. ACL 2025. arXiv:2506.01048
  • Sun Y., Xie E., Feng T., You J. GMTRouter: Personalized LLM Router over Multi-turn User Interactions. EMNLP 2026 Findings. arXiv:2511.08590
  • ProbeDirichlet / RouterXBench. 2026. arXiv:2602.11877

Online / bandits

  • Wang X. et al. MixLLM: Dynamic Routing in Mixed Large Language Models. NAACL 2025. arXiv:2502.18482
  • Dai X., Li J., Liu X., Yu A., Lui J.C.S. Cost-Effective Online Multi-LLM Selection with Versatile Reward Models (C2MAB-V). 2024. arXiv:2405.16587
  • Li L., Chu W., Langford J., Schapire R.E. A Contextual-Bandit Approach to Personalized News Article Recommendation. WWW 2010
  • Auer P., Cesa-Bianchi N., Fischer P. Finite-time Analysis of the Multiarmed Bandit Problem. Machine Learning 47, 2002
  • Thompson W.R. On the Likelihood that One Unknown Probability Exceeds Another. Biometrika 1933
  • Bao Z. et al. OrcaRouter: A Production-Oriented LLM Router with Hybrid Offline-Online Learning. 2026. arXiv:2605.30736
  • Drift-Aware Sparse Routing: Nonstationary Contextual Routing under Multiple Knapsack Constraints. 2026. arXiv:2609.00662
  • Raj R. et al. TRACE-Router: Task-Consistent and Adaptive Online Routing for Agentic AI. 2026. arXiv:2607.22465
  • Federate the Router. 2026. arXiv:2601.22318
  • Patel S., Parthasarathy A.R., Mallick A., Joshi G. Beyond Accuracy and Cost: Latency-Aware LLM Query Routing for Dynamic Workloads. 2026. arXiv:2607.18253
  • HW-Router. DAC 2026. arXiv:2608.14575
  • Garivier A., Moulines E. On Upper-Confidence Bound Policies for Switching Bandit Problems. ALT 2011
  • Joulani P., György A., Szepesvári C. Online Learning under Delayed Feedback. ICML 2013
  • Badanidiyuru A., Kleinberg R., Slivkins A. Bandits with Knapsacks. FOCS 2013
  • Yue Y., Broder J., Kleinberg R., Joachims T. The K-armed Dueling Bandits Problem. JCSS 2012
  • Drugan M.M., Nowé A. Designing Multi-Objective Multi-Armed Bandits Algorithms. IJCNN 2013
  • Dudík M., Langford J., Li L. Doubly Robust Policy Evaluation and Learning. ICML 2011

Cascades & multi-round

  • Chen L., Zaharia M., Zou J. FrugalGPT. 2023. arXiv:2305.05176
  • Aggarwal P. et al. AutoMix: Automatically Mixing Language Models. NeurIPS 2024. arXiv:2310.12963
  • Zhang H., Feng T., You J. Router-R1: Teaching LLMs Multi-Round Routing and Aggregation via RL. NeurIPS 2025. arXiv:2506.09033
  • Jiang D., Ren X., Lin B.Y. LLM-Blender. ACL 2023. arXiv:2306.02561
  • Dekoninck J., Baader M., Vechev M. A Unified Approach to Routing and Cascading for LLMs. 2024. arXiv:2410.10347
  • Huang S., Wang S., Ma H., Xu Z. RLCascadeRouter: Quality-Estimator-Free Cascade Routing via Reinforcement Learning. 2026. arXiv:2608.15817
  • Shaikh N. Knowing When to Ask for Help: Bayesian Self-Escalation in Hierarchical LLM Agents. 2026. arXiv:2608.24087
  • Lei J., Huang J. TACIT-Switch: Cost-Aware Model Escalation for LLM Agents from Censored Supervision. 2026. arXiv:2608.27911
  • Li S., Abdelmoniem A.M., Wang S. ProgRouter: Online Progress-Guided Orchestration for Multi-Agent LLM Workflows. EMNLP 2026 Findings. arXiv:2608.25992
  • MTRouter: Multi-Turn LLM Routing. ACL 2026. arXiv:2604.23530
  • Tang S. et al. Beyond Gemini-3-Pro: Revisiting LLM Routing and Aggregation at Scale (JiSi). 2026. arXiv:2601.01330
  • Confidence-Guided Stepwise Model Routing. 2025. arXiv:2511.06190
  • Differential Reasoning Router. 2026. arXiv:2608.30224

Risk control, calibration, deferral

  • RACER: Risk-Controlled LLM Routing with Abstention. 2026. arXiv:2603.06616
  • RouteNLP: Conformal Cascade Thresholds and Distillation–Routing Co-optimisation. ACL 2026 Industry. arXiv:2604.23577
  • CR²: Cost-Aware Risk-Controlled Routing. 2026. arXiv:2605.12001
  • Guo C., Pleiss G., Sun Y., Weinberger K.Q. On Calibration of Modern Neural Networks. ICML 2017
  • Kuhn L., Gal Y., Farquhar S. Semantic Uncertainty. ICLR 2023 · Kadavath S. et al. Language Models (Mostly) Know What They Know. 2022
  • Madras D., Pitassi T., Zemel R. Predict Responsibly: Improving Fairness and Accuracy by Learning to Defer. NeurIPS 2018
  • Mozannar H., Sontag D. Consistent Estimators for Learning to Defer to an Expert. ICML 2020
  • Verma R., Barrejón D., Nalisnick E. Learning to Defer to Multiple Experts. AISTATS 2023

Agents, tools, skills

  • Yue Y. et al. MasRouter: Learning to Route LLMs for Multi-Agent Systems. 2025. arXiv:2502.11133
  • Qin Y. et al. ToolLLM: Facilitating LLMs to Master 16000+ Real-world APIs. 2023. arXiv:2307.16789
  • Du Y., Wei F., Zhang H. AnyTool: Self-Reflective, Hierarchical Agents for Large-Scale API Calls. 2024. arXiv:2402.04253
  • Enrich-Retrieve-Rank: Discovery for Models, Agents, Tools and Skills at Scale. 2026. arXiv:2608.22695
  • SCOUT: Scalable Tool Discovery for MCP Gateways. PayPal, 2026. arXiv:2608.23992
  • Task2MCP / T2MRec: MCP Server Recommendation. 2026. arXiv:2604.17234
  • Semantic Tool Discovery for MCP. 2026. arXiv:2603.20313 · SchemaRouter. arXiv:2608.21375 · CacheRouter. arXiv:2608.22708 · Nexus. arXiv:2608.20397 · ToolDNS. arXiv:2607.18242 · ComplexMCP. arXiv:2605.10787
  • Best Prefix Selection: Optimal Skill Selection under Token Budgets. 2026. arXiv:2608.19993
  • SkillFeed: Personalized Skill Routing. 2026. arXiv:2608.28241
  • Compositional Skill Routing. arXiv:2606.18051 · CaSKG. arXiv:2608.25500 · A Systems Foundation for Agentic Skills. arXiv:2608.29596 · Microskill Architecture. arXiv:2606.05720
  • Agent-as-a-Router. arXiv:2606.22902 · SWE-Router. arXiv:2607.00053 · Agentic Routing: a harness-native data flywheel. arXiv:2607.11399 · Routing Is Least Learnable Where It Is Most Valuable. arXiv:2608.06171 · LLMs Predict Failure Risk but Not Which Collaboration Protocol Pays Off. arXiv:2608.14927
  • BudgetMem. ICML 2026. arXiv:2602.06025 · Gated-Memory Routing. arXiv:2609.00237
  • LatentRouter. arXiv:2605.11301 · VLM Routing for Computer-Use Agents. arXiv:2603.12823 · Modality Escalation. arXiv:2607.05438

Economics & mechanism design

  • Chen H. et al. Error-Aware Reverse Auction Mechanism for Large Language Model Routing (EA-RAM). 2026. arXiv:2608.12719
  • Strategy Auctions for Small Agents. ICML 2026. arXiv:2602.02751
  • Coalition Formation and Communication Pricing in Skill-Based Agentic AI. 2026. arXiv:2608.07532
  • Ghodsi A. et al. Dominant Resource Fairness. NSDI 2011

Sequential decisions & queueing

  • Levin E., Pieraccini R., Eckert W. A Stochastic Model of Human-Machine Interaction for Learning Dialog Strategies. IEEE TSAP 2000
  • Young S., Gašić M., Thomson B., Williams J.D. POMDP-Based Statistical Spoken Dialog Systems: A Review. Proc. IEEE 2013
  • Koole G., Mandelbaum A. Queueing Models of Call Centers: An Introduction. Annals of OR 2002
  • Kingman J.F.C. The Single Server Queue in Heavy Traffic. 1961 · Little J.D.C. A Proof for the Queuing Formula L = λW. 1961

Security

  • Shafran A., Schuster R., Ristenpart T., Shmatikov V. Rerouting LLM Routers. 2025. arXiv:2501.01818
  • ROPE: Routed Origin Policy Enforcement against Indirect Prompt Injection. 2026. arXiv:2608.27496
  • Verifiable Manifest Signing for MCP Pipelines. 2026. arXiv:2601.23132
  • Recognition Without Enforcement. 2026. arXiv:2608.28502
  • Beyond Max Tokens: Resource Amplification via Tool Chains. 2026. arXiv:2601.10955
  • When Safety Routing Breaks. 2026. arXiv:2609.01455
  • Malicious Prompt Detection via Bootstrap Aggregation. 2026. arXiv:2602.08062

Benchmarks

  • Hu Q.J. et al. RouterBench: A Benchmark for Multi-LLM Routing System. 2024. arXiv:2403.12031
  • Huang Z. et al. RouterEval: A Comprehensive Benchmark for Routing LLMs. 2025. arXiv:2503.10657
  • LLMRouterBench: A Unified Benchmark for LLM Routing. 2026. arXiv:2601.07206
  • Feng T., You J. et al. LLMRouter / xRouteBench: Routing as Sequential Decision Making. 2026. arXiv:2608.06867
  • RouterXBench. 2026. arXiv:2602.11877
  • How Much of the Routing Gap Is Real? Label Noise and the Oracle. 2026. arXiv:2607.03436
  • Most of the LLM Routing Gap Is Task Type. 2026. arXiv:2608.23023
  • When is Routing Meaningful? Diversity and Perturbation Robustness. 2026. arXiv:2607.09197
  • Characterising Energy per Request and per Token for LLM Serving. IISWC 2026. arXiv:2608.28044

Drift

  • Page E.S. Continuous Inspection Schemes. Biometrika 1954 · Hinkley D.V. 1971
  • Bifet A., Gavaldà R. Learning from Time-Changing Data with Adaptive Windowing (ADWIN). SDM 2007

Systems

  • vLLM Project. vLLM Semantic Router. github.com/vllm-project/semantic-router · technical report arXiv:2603.04444
  • LMSYS. RouteLLM. github.com/lm-sys/RouteLLM
  • ulab-uiuc. LLMRouter. github.com/ulab-uiuc/LLMRouter · GMTRouter. github.com/ulab-uiuc/GMTRouter
  • Generating Workflow DAGs from Natural-Language Routing Rules. 2026. arXiv:2608.30250