--- title: Claude Code burns tokens like it's trying to win a... description: Claude Code is consuming significantly more tokens than competing AI coding agents, with reports of 2x to 5.6x higher token usage and one comparison showing 9x... canonical: https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Claude Code burns tokens like it's trying to win a benchmark, not save your money | daily.dev og:description: Claude Code is consuming significantly more tokens than competing AI coding agents, with reports of 2x to 5.6x higher token usage and one comparison showing 9x... og:url: https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig og:image: https://api.daily.dev/og/posts/KLORPOviG.png og:image:alt: Claude Code burns tokens like it's trying to win a benchmark, not save your money 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. # Claude Code burns tokens like it's trying to win a benchmark, not save your money **[Trends](https://daily.dev/sources/trends)** · 2 min read · 1 upvotes · 1 comments ## Summary Claude Code is consuming significantly more tokens than competing AI coding agents, with reports of 2x to 5.6x higher token usage and one comparison showing 9x higher costs per task versus Kimi Code. Possible explanations include unoptimized context management, a bug, or a deliberate design choice where verbosity improves results on complex tasks. The core problem is a lack of transparency from Anthropic — users are hitting plan ceilings and watching API costs balloon without clear explanation. No clean apples-to-apples benchmark yet exists that controls for both cost and task success rate simultaneously, leaving the debate unresolved but the token appetite undeniable. ## Content The discourse around Claude Code right now isn't about capability. It's about cost, and the numbers being thrown around are genuinely alarming. One user on the $100/month Claude Code Max plan burned through three 5-hour windows — roughly $600 in tokens at retail pricing. Another comparison puts Claude Code at 5.6x more tokens and 9x more cost per task than Kimi Code. A separate researcher found it uses 2-3x more tokens than other harnesses at similar success rates, and openly wonders whether that's a bug, a design choice, or just sloppiness. The most detailed breakdown comes from a 30-day trace across a 45-person engineering team. The finding: only 14% of input tokens are actual user prompts. The rest is overhead — static system prompts, MCP server schemas, CLAUDE.md files, tool results, prior assistant context, and thinking tokens. You're not paying for your work. You're paying for Claude Code's scaffolding. Sub-agents make this worse. Multi-agent workflows consume 4-7x more tokens than single-agent sessions because each subagent spins up in an isolated context with no memory of what the main session already established. So the system re-explains itself, repeatedly, at your expense. And you never see the subagent's internal reasoning — just a summary — so you can't even audit where the tokens went. The open question is whether this is intentional. The researcher tracking token efficiency across harnesses notes that higher token usage does seem to correlate with better performance on harder tasks. So Anthropic may be making a deliberate tradeoff: burn more tokens, get better results, let users figure out the economics. That's a defensible product decision. It's also one that Anthropic hasn't been particularly transparent about. Practical mitigations exist: strip unused MCP servers, keep CLAUDE.md short, use medium thinking effort instead of max, end sessions early, and default to Sonnet over Opus for routine work. Sub-agents should be reserved for genuinely parallel workloads, not convenience. But the underlying tension is real. Claude Code is positioned as a professional tool for serious engineering work. At these token rates, the economics only work if the output quality justifies the spend — and right now, a lot of developers are running the math and coming up short. ## Community discussion Top comments from developers on daily.dev. **@randy** · 0 upvotes > Wowzers! If that pricing is accurate, and what I see in OpenUsage is accurate, I've done a _lot_ of work in the last 30 days. `gpt-5.6-sol` on medium and prior to that `gpt-5.5` on high... Cost is under $500. ## Similar posts on daily.dev - [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 --- Tags: [#ai-coding](https://daily.dev/tags/ai-coding), [#anthropic](https://daily.dev/tags/anthropic), [#claude-code](https://daily.dev/tags/claude-code), [#finops](https://daily.dev/tags/finops) [View this post on daily.dev](https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig) ```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":"Claude Code burns tokens like it's trying to win a benchmark, not save your money","url":"https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig"},"datePublished":"2026-07-30T23:13:49.185Z","dateModified":"2026-07-31T14:38:24.230Z","description":"Claude Code is consuming significantly more tokens than competing AI coding agents, with reports of 2x to 5.6x higher token usage and one comparison showing 9x...","image":"https://pbs.twimg.com/media/HOgwLo8XAAABETp.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HOgwLo8XAAABETp.jpg","isAccessibleForFree":true,"articleSection":"Trends","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":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":1,"discussionUrl":"https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"keywords":"ai-coding,anthropic,claude-code,finops","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Claude Code burns tokens like it's trying to win a benchmark, not save your money"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/claude-code-burns-tokens-like-it-s-trying-to-win-a-benchmark-not-save-your-money-klorpovig","comment":[{"@type":"Comment","text":"Wowzers! If that pricing is accurate, and what I see in OpenUsage is accurate, I’ve done a lot of work in the last 30 days. gpt-5.6-sol on medium and prior to that gpt-5.5 on high… Cost is under $500.","datePublished":"2026-07-31T13:27:59.002Z","url":"https://daily.dev/posts/KLORPOviG#c-n44AJiEcw","author":{"@type":"Person","name":"Randy","url":"https://daily.dev/randy","image":"https://media.daily.dev/image/upload/s--UjV4-KkB--/f_auto/v1708097210/avatars/avatar_HXYbbGcBO38Rfv7RrCBdA"}}]} ```