--- title: I set up Claude Code the way Spotify does to save... description: An experiment replicates Spotify's open-source shunt plugin, which offloads cheap read-heavy tasks from Claude Code to a cheaper worker model to cut token... canonical: https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: I set up Claude Code the way Spotify does to save tokens, and it cost me more | daily.dev og:description: An experiment replicates Spotify's open-source shunt plugin, which offloads cheap read-heavy tasks from Claude Code to a cheaper worker model to cut token... og:url: https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy og:image: https://api.daily.dev/og/posts/DC3aknhmY.png og:image:alt: I set up Claude Code the way Spotify does to save tokens, and it cost me more 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. # I set up Claude Code the way Spotify does to save tokens, and it cost me more **[XDA Developers](https://daily.dev/sources/xda-developers)** · 5 min read · 0 upvotes · 0 comments ## Summary An experiment replicates Spotify's open-source shunt plugin, which offloads cheap read-heavy tasks from Claude Code to a cheaper worker model to cut token costs, reportedly saving up to 90% in Spotify's own Java monorepo benchmarks. Testing the plugin against Opus 5.5 with Qwen3-Coder-30B-A3B (local, via Lemonade Server) and Claude Haiku as workers showed delegation actually used more tokens overall (23% more with Qwen3-Coder, 34% more with Haiku) and took 7-10x longer than Opus working alone, because Opus already reads files efficiently via grep rather than full reads, the plugin's hooks miss many of Opus's actual tool calls, and Opus still has to re-check or rewrite delegated drafts that miss spec requirements. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Does delegating Claude Code tasks to a cheaper worker model like Spotify's shunt plugin actually save tokens with Opus 5.5? No, in practice it used more tokens, not fewer. Opus 5.5 working alone used 31,900 tokens, while delegating to local Qwen3-Coder-30B-A3B used 39,300 tokens (23% more) and delegating to Claude Haiku used 42,800 tokens (34% more), because Opus still had to read and often rewrite the worker's flawed draft to pass tests. _Anyone weighing agentic delegation setups against plain Opus usage can compare real token and cost trade-offs on daily.dev._ ### What token savings does Spotify's shunt plugin achieve on large file reads? Spotify's shunt README reports 82% to 94% token savings on large reads against a 162,000-line Java monorepo, with a mean of 90%, using Gemini 2.5 Flash as the worker model behind Spotify Portal, its commercial developer platform that requires contacting sales for access. _Developers evaluating token-saving plugins for large codebases can dig into benchmarks like this on daily.dev._ ### Why did a local Qwen3-Coder model fail when used as a worker for Claude Code delegation? On Lemonade Server's ROCm backend, Qwen3-Coder-30B-A3B invented function names that did not exist and failed to copy file content correctly, while the same model on the Vulkan backend copied lines character for character, indicating a backend-specific reliability issue rather than a model limitation. It also ran at roughly 12 tokens per second, causing timeouts on delegated tasks. _Developers debugging local LLM backend quirks can track similar hands-on findings on daily.dev._ ## Similar posts on daily.dev - [Spotify’s Backstage Portal cut my Claude Code token usage by 90%](https://daily.dev/posts/spotify-s-backstage-portal-cut-my-claude-code-token-usage-by-90--2beo3hwm6) · Spotify Labs · 4 upvotes · 1 comments - [Claude Code Sends 4.7x More Tokens Than OpenCode Before Reading Your Prompt](https://daily.dev/posts/claude-code-sends-4-7x-more-tokens-than-opencode-before-reading-your-prompt-9m02iom1z) · Hacker News · 6 upvotes · 0 comments - [7 ways I make Claude Code use fewer tokens without hurting the results](https://daily.dev/posts/7-ways-i-make-claude-code-use-fewer-tokens-without-hurting-the-results-s1mmfwq7r) · XDA Developers · 1 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#anthropic](https://daily.dev/tags/anthropic), [#claude-code](https://daily.dev/tags/claude-code), [#local-ai](https://daily.dev/tags/local-ai), [#context-engineering](https://daily.dev/tags/context-engineering) [View this post on daily.dev](https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy) ```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":"I set up Claude Code the way Spotify does to save tokens, and it cost me more","url":"https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy"},"datePublished":"2026-10-03T23:00:25.658Z","dateModified":"2026-10-05T19:37:57.340Z","description":"An experiment replicates Spotify's open-source shunt plugin, which offloads cheap read-heavy tasks from Claude Code to a cheaper worker model to cut token...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0710cdbaeebe6d8bc0c225e05894c111?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0710cdbaeebe6d8bc0c225e05894c111?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"XDA Developers","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":"XDA Developers","logo":"https://media.daily.dev/image/upload/s--kCg6yyAP--/f_auto,q_auto/v1774964407/logos/xda-developers?_a=BAMAMiWQ0","url":"https://daily.dev/sources/xda-developers"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,anthropic,claude-code,local-ai,context-engineering","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"XDA Developers","item":"https://daily.dev/sources/xda-developers"},{"@type":"ListItem","position":3,"name":"I set up Claude Code the way Spotify does to save tokens, and it cost me more"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/i-set-up-claude-code-the-way-spotify-does-to-save-tokens-and-it-cost-me-more-dc3aknhmy#faq","mainEntity":[{"@type":"Question","name":"Does delegating Claude Code tasks to a cheaper worker model like Spotify's shunt plugin actually save tokens with Opus 5.5?","acceptedAnswer":{"@type":"Answer","text":"No, in practice it used more tokens, not fewer. Opus 5.5 working alone used 31,900 tokens, while delegating to local Qwen3-Coder-30B-A3B used 39,300 tokens (23% more) and delegating to Claude Haiku used 42,800 tokens (34% more), because Opus still had to read and often rewrite the worker's flawed draft to pass tests. Anyone weighing agentic delegation setups against plain Opus usage can compare real token and cost trade-offs on daily.dev."}},{"@type":"Question","name":"What token savings does Spotify's shunt plugin achieve on large file reads?","acceptedAnswer":{"@type":"Answer","text":"Spotify's shunt README reports 82% to 94% token savings on large reads against a 162,000-line Java monorepo, with a mean of 90%, using Gemini 2.5 Flash as the worker model behind Spotify Portal, its commercial developer platform that requires contacting sales for access. Developers evaluating token-saving plugins for large codebases can dig into benchmarks like this on daily.dev."}},{"@type":"Question","name":"Why did a local Qwen3-Coder model fail when used as a worker for Claude Code delegation?","acceptedAnswer":{"@type":"Answer","text":"On Lemonade Server's ROCm backend, Qwen3-Coder-30B-A3B invented function names that did not exist and failed to copy file content correctly, while the same model on the Vulkan backend copied lines character for character, indicating a backend-specific reliability issue rather than a model limitation. It also ran at roughly 12 tokens per second, causing timeouts on delegated tasks. Developers debugging local LLM backend quirks can track similar hands-on findings on daily.dev."}}]} ```