--- title: Preferences Over Benchmarks: Model Routing — Archana... description: DigitalOcean engineers present the case against relying on a single frontier LLM for all tasks, arguing that cost, task fit, and reliability risk push teams... canonical: https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean | daily.dev og:description: DigitalOcean engineers present the case against relying on a single frontier LLM for all tasks, arguing that cost, task fit, and reliability risk push teams... og:url: https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g og:image: https://api.daily.dev/og/posts/Z44jGaF0g.png og:image:alt: Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean **[AI Engineer](https://daily.dev/sources/aidotengineer)** · 15 min read · 2 upvotes · 0 comments ## Summary DigitalOcean engineers present the case against relying on a single frontier LLM for all tasks, arguing that cost, task fit, and reliability risk push teams toward multi-model routing. They introduce DigitalOcean's open-source inference router, which uses a purpose-built, sub-200ms routing model to pick the right model per request based on user-defined preferences (cost, latency, quality, task type) rather than public benchmarks. A live demo compares a single-model Claude Opus workflow against the router inside a coding agent, showing similar output quality at roughly a third of the cost across multiple coding tasks (code generation, optimization, unit tests, documentation). The router is free, requires no code changes, and builds toward evals, caching, and personalization as future layers. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does DigitalOcean's inference router decide which LLM to use for a given request? It uses a custom mixture-of-experts model purpose-built for routing, which reads a task description plus user-defined preferences (cost, latency, quality, preferred models, or hard rules) and picks a model per request in under 200 milliseconds. It runs through an open source proxy with no vendor lock-in and costs nothing extra to use, and DigitalOcean open sourced the routing model via a project called Plano. _Developers weighing model-routing setups can track releases like this router on daily.dev._ ### How much can model routing reduce LLM costs compared to always using a premium model like Claude Opus? In a live demo comparing a coding agent workflow, a software-engineering router spent 14 cents total across several coding tasks versus 44 cents when using Claude Opus directly for every request, roughly a 3x cost reduction with similar output quality. Individual steps showed similar ratios, such as 8 cents versus 25 cents for one feature-building task. _Teams comparing model costs for coding agents can follow benchmarks like this on daily.dev._ ### Why might a single best LLM model not exist for all coding tasks? Model fit depends on the specific task: classification and labeling can run well on small open models for cost savings, in-line code completion needs a fast routing-optimized model, code generation and bug fixing work well with mid-sized open-weight models, and accuracy-critical work like code review or security tasks calls for a frontier model. Cost, latency needs, and end-user preference also shape which model fits each request. _Developers deciding between model tiers for different coding tasks can track this reasoning on daily.dev._ ## Similar posts on daily.dev - [DigitalOcean Inference Router Enhancements: Performance Tuned Routing, No Rules to Write](https://daily.dev/posts/digitalocean-inference-router-enhancements-performance-tuned-routing-no-rules-to-write-duiodam0z) · DigitalOcean · 0 upvotes · 0 comments - [Why Your Best Model Is Two Models: Routing Between Kimi K3 and Claude](https://daily.dev/posts/why-your-best-model-is-two-models-routing-between-kimi-k3-and-claude-ejhl4kv4j) · DigitalOcean Community · 0 upvotes · 0 comments - [LLM Routing Can Cost More Than Not Routing](https://daily.dev/posts/llm-routing-can-cost-more-than-not-routing-hujhqshkk) · Daily Dose of Data Science \| Avi Chawla \| Substack · 1 upvotes · 0 comments - [Multi-Model Routing Is an Infrastructure Decision, Not a Feature: The Tradeoffs Nobody Publishes](https://daily.dev/posts/multi-model-routing-is-an-infrastructure-decision-not-a-feature-the-tradeoffs-nobody-publishes-zbm50wrn5) · DigitalOcean Community · 0 upvotes · 0 comments - [Model Evaluations: Prove Your Routing Policy Actually Works](https://daily.dev/posts/model-evaluations-prove-your-routing-policy-actually-works-nquasumiq) · DigitalOcean · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#finops](https://daily.dev/tags/finops), [#ai-inference](https://daily.dev/tags/ai-inference), [#digitalocean](https://daily.dev/tags/digitalocean), [#ai-gateway](https://daily.dev/tags/ai-gateway) [View this post on daily.dev](https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean","url":"https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g"},"datePublished":"2026-08-22T15:38:35.101Z","dateModified":"2026-09-14T07:43:51.936Z","description":"DigitalOcean engineers present the case against relying on a single frontier LLM for all tasks, arguing that cost, task fit, and reliability risk push teams...","image":"https://i.ytimg.com/vi/FvxY8oPoI8o/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/FvxY8oPoI8o/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AI Engineer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"AI Engineer","logo":"https://media.daily.dev/image/upload/s--u5PucxNT--/f_auto/v1724338940/logos/aidotengineer","url":"https://daily.dev/sources/aidotengineer"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,finops,ai-inference,digitalocean,ai-gateway","timeRequired":"PT15M","video":{"@type":"VideoObject","name":"Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean","description":"DigitalOcean engineers present the case against relying on a single frontier LLM for all tasks, arguing that cost, task fit, and reliability risk push teams...","thumbnailUrl":"https://i.ytimg.com/vi/FvxY8oPoI8o/sddefault.jpg","uploadDate":"2026-08-22T15:38:35.101Z","duration":"PT15M","url":"https://api.daily.dev/r/Z44jGaF0g","embedUrl":"https://www.youtube.com/embed/FvxY8oPoI8o"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AI Engineer","item":"https://daily.dev/sources/aidotengineer"},{"@type":"ListItem","position":3,"name":"Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/preferences-over-benchmarks-model-routing-archana-kamath-tyler-gillam-digitalocean-z44jgaf0g#faq","mainEntity":[{"@type":"Question","name":"How does DigitalOcean's inference router decide which LLM to use for a given request?","acceptedAnswer":{"@type":"Answer","text":"It uses a custom mixture-of-experts model purpose-built for routing, which reads a task description plus user-defined preferences (cost, latency, quality, preferred models, or hard rules) and picks a model per request in under 200 milliseconds. It runs through an open source proxy with no vendor lock-in and costs nothing extra to use, and DigitalOcean open sourced the routing model via a project called Plano. Developers weighing model-routing setups can track releases like this router on daily.dev."}},{"@type":"Question","name":"How much can model routing reduce LLM costs compared to always using a premium model like Claude Opus?","acceptedAnswer":{"@type":"Answer","text":"In a live demo comparing a coding agent workflow, a software-engineering router spent 14 cents total across several coding tasks versus 44 cents when using Claude Opus directly for every request, roughly a 3x cost reduction with similar output quality. Individual steps showed similar ratios, such as 8 cents versus 25 cents for one feature-building task. Teams comparing model costs for coding agents can follow benchmarks like this on daily.dev."}},{"@type":"Question","name":"Why might a single best LLM model not exist for all coding tasks?","acceptedAnswer":{"@type":"Answer","text":"Model fit depends on the specific task: classification and labeling can run well on small open models for cost savings, in-line code completion needs a fast routing-optimized model, code generation and bug fixing work well with mid-sized open-weight models, and accuracy-critical work like code review or security tasks calls for a frontier model. Cost, latency needs, and end-user preference also shape which model fits each request. Developers deciding between model tiers for different coding tasks can track this reasoning on daily.dev."}}]} ```