--- title: Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS... description: Pi, a minimal AI coding agent harness, is argued to be the best way to run any AI model — especially cheaper open models like DeepSeek. The core argument... canonical: https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST! | daily.dev og:description: Pi, a minimal AI coding agent harness, is argued to be the best way to run any AI model — especially cheaper open models like DeepSeek. The core argument... og:url: https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev og:image: https://api.daily.dev/og/posts/GHW6l2veV.png og:image:alt: Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST! 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. # Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST! **[AICodeKing](https://daily.dev/sources/aicodeking)** · 8 min read · 0 upvotes · 0 comments ## Summary Pi, a minimal AI coding agent harness, is argued to be the best way to run any AI model — especially cheaper open models like DeepSeek. The core argument centers on Pi's tiny system prompt (a few hundred tokens vs. Claude Code's ~10,000), which makes it model-agnostic and avoids confusing models tuned for other harnesses. Anthropic recently cut Claude Code's system prompt by ~80% with no performance loss, validating the minimal-prompt philosophy. Composio benchmark data across 30 agentic tasks shows Oh My Pi (a Pi fork) beating Claude Code, Codex, and Open Code on tasks completed (17/30) when using DeepSeek V4 Flash, while Claude Code was fastest but nearly 3x more expensive per successful task. The video also walks through Pi's features: clean terminal UI, 15+ model providers, mid-session model switching with preserved context, a session tree for branching conversations, an extension system, and headless/CI modes. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How did Pi agent compare to Claude Code and Codex on the Composio benchmark with DeepSeek? Oh My Pi (a Pi-based harness) completed 17 out of 30 agentic tasks, beating Claude Code and Codex (both 16) and Open Code (14) when running DeepSeek V4 Flash. Claude Code was fastest at 123 seconds median per task but cost nearly 20 cents per successful task — almost 3x more than Open Code at 7 cents. Oh My Pi cost about 10 cents per successful task. _Developers choosing between agent harnesses for DeepSeek workloads track cost-vs-performance tradeoffs like these on daily.dev._ ### Why does a smaller system prompt make an AI coding agent work better with open models like DeepSeek? Large system prompts are tuned for a specific model — Claude Code's ~10,000-token prompt was written for Claude. When you run DeepSeek or other open models through it, the model processes thousands of tokens of instructions never designed for it, which can actively confuse it and adds cost to every request. A minimal prompt (a few hundred tokens) lets the model's own capabilities drive behavior, making the harness model-agnostic. _Teams running open models in production find harness selection discussions like this on daily.dev._ ## Similar posts on daily.dev - [My favorite AI coding tool isn't OpenCode: it's something way better \(and cooler\)](https://daily.dev/posts/my-favorite-ai-coding-tool-isn-t-opencode-it-s-something-way-better-and-cooler--mhoh5tl9y) · XDA Developers · 0 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#ai-agents](https://daily.dev/tags/ai-agents), [#claude-code](https://daily.dev/tags/claude-code), [#deepseek](https://daily.dev/tags/deepseek) [View this post on daily.dev](https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev) ```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":"Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST!","url":"https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev"},"datePublished":"2026-08-09T09:36:31.343Z","dateModified":"2026-08-11T02:12:11.423Z","description":"Pi, a minimal AI coding agent harness, is argued to be the best way to run any AI model — especially cheaper open models like DeepSeek. The core argument...","image":"https://i.ytimg.com/vi/WoOWe1SrIQo/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/WoOWe1SrIQo/sddefault.jpg","isAccessibleForFree":true,"articleSection":"AICodeKing","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":"AICodeKing","logo":"https://media.daily.dev/image/upload/s--x7nDUfWj--/f_auto,q_auto/v1768208312/logos/aicodeking","url":"https://daily.dev/sources/aicodeking"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,ai-agents,claude-code,deepseek","timeRequired":"PT8M","video":{"@type":"VideoObject","name":"Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST!","description":"Pi, a minimal AI coding agent harness, is argued to be the best way to run any AI model — especially cheaper open models like DeepSeek. The core argument...","thumbnailUrl":"https://i.ytimg.com/vi/WoOWe1SrIQo/sddefault.jpg","uploadDate":"2026-08-09T09:36:31.343Z","duration":"PT8M","url":"https://api.daily.dev/r/GHW6l2veV","embedUrl":"https://www.youtube.com/embed/WoOWe1SrIQo"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"AICodeKing","item":"https://daily.dev/sources/aicodeking"},{"@type":"ListItem","position":3,"name":"Pi Agent + GPT-5.6 Luna, V4 Flash & Every Model: THIS IS THE BEST!"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/pi-agent-gpt-5-6-luna-v4-flash-every-model-this-is-the-best--ghw6l2vev#faq","mainEntity":[{"@type":"Question","name":"How did Pi agent compare to Claude Code and Codex on the Composio benchmark with DeepSeek?","acceptedAnswer":{"@type":"Answer","text":"Oh My Pi (a Pi-based harness) completed 17 out of 30 agentic tasks, beating Claude Code and Codex (both 16) and Open Code (14) when running DeepSeek V4 Flash. Claude Code was fastest at 123 seconds median per task but cost nearly 20 cents per successful task — almost 3x more than Open Code at 7 cents. Oh My Pi cost about 10 cents per successful task. Developers choosing between agent harnesses for DeepSeek workloads track cost-vs-performance tradeoffs like these on daily.dev."}},{"@type":"Question","name":"Why does a smaller system prompt make an AI coding agent work better with open models like DeepSeek?","acceptedAnswer":{"@type":"Answer","text":"Large system prompts are tuned for a specific model — Claude Code's ~10,000-token prompt was written for Claude. When you run DeepSeek or other open models through it, the model processes thousands of tokens of instructions never designed for it, which can actively confuse it and adds cost to every request. A minimal prompt (a few hundred tokens) lets the model's own capabilities drive behavior, making the harness model-agnostic. Teams running open models in production find harness selection discussions like this on daily.dev."}}]} ```