Routing & Cascades

Routing & Cascades: Classifiers, verifiers, cascades, and the benchmarks that judge them. 42 field notes from the engineers shipping Nadir's LLM routing layer.

Posts in Routing & Cascades

• We Beat Claude Opus 5 With Claude Opus 5 (2026-07-30)
• LLM Routing Explained: How AI Model Routing Cuts Costs (2026-07-14)
• The Blanket Swap (2026-07-09)
• FrugalGPT Cascade Routing: A Step-by-Step Implementation (2026-06-29)
• Nadir vs Levanto Sage 2: AI Agent Routing Benchmark (2026-11-03)
• The Replan Rate (2026-11-02)
• Route Like a Bank (2026-11-02)
• The Default Is the Bill (2026-10-23)
• The 98% Demotion (2026-10-19)
• The Ten-Point Discount (2026-10-18)
• The 71/23 Split (2026-10-16)
• After the Answer (2026-10-13)
• 95% Easy, 5% Saved (2026-10-07)
• Six Deciders, Sixteen Days (2026-10-04)
• Where the Label Lives (2026-10-01)
• The $0.0002 Judge (2026-09-29)
• Three Down at Once (2026-09-29)
• Buy the Hint (2026-09-28)
• Jev, Laya, GLiNER: Beaten by len() (2026-09-28)
• When LLM Routing Does Not Help (2026-09-28)
• Right Answer, Wrong Shape (2026-09-26)
• Jev Decision Model: Is It Really 400x Cheaper? (2026-09-23)
• The Routing Tax (2026-09-21)
• Ask The Weights (2026-09-19)
• The One-Sided Signal (2026-09-08)
• Confidently Wrong (2026-09-03)
• The Threshold Nobody Tunes (2026-08-31)
• The Verification Discount (2026-08-30)
• The Retry Tax (2026-08-24)
• The Classification Ceiling (2026-08-22)
• RouterArena Benchmark: How Far LLM Routers Fall From the Oracle (2026-08-17)
• The Router in the Middle (2026-08-07)
• Shadow Mode, Forever (2026-07-20)
• New Model, Same Bar (2026-07-18)
• The Guardrail Tax (2026-07-06)
• The Router That Refuses to Guess (2026-07-02)
• The 7B Substitution (2026-07-01)
• LLMRouterBench: Most LLM Routers Fail to Beat a Simple Baseline (2026-06-29)
• RouteLLM: 74% of Queries Sent to a Cheap Model at 95% Quality (2026-06-27)
• Two Hybrid Routers, One Pattern (2026-06-10)
• Routing without verification is dead-reckoning (2026-05-27)
• Reference-assisted RouterBench cascade: the 60% / 98% research ceiling (2026-05-26)

Pages on this site

• Home
• Pricing
• Nadir Lens
• Benchmarks
• Techniques
• What is an LLM router?
• AI gateway vs model router
• Savings calculator
• Solutions
• Routing
• Provider fallback
• Analytics
• Prompt clustering
• Compare
• Nadir vs LiteLLM
• Nadir vs OpenRouter
• Nadir vs Requesty
• Nadir vs Ramp Router
• Nadir vs Not Diamond
• Nadir vs Weave Router
• Nadir vs Portkey
• Contact
• Self-host with NadirClaw
• On-prem
• Context Optimize
• Docs
• Blog
• Agent skill
• Claude Code receipt
• Coding agents
• Cookbooks
• FAQ
• How it works
• Terms of Service
• Privacy Policy
• Data Processing Agreement
• Security

Machine-readable summaries of this site: llms.txt and llms-full.txt.