[{"data":1,"prerenderedAt":251},["ShallowReactive",2],{"blog-pi-agent":3},{"id":4,"title":5,"body":6,"date":233,"description":234,"extension":235,"featured":236,"meta":237,"navigation":238,"part":239,"path":240,"seo":241,"series":242,"stem":243,"tags":244,"tldr":249,"__hash__":250},"blog\u002Fblog\u002Fpi-agent.md","The Real Cost of AI Coding Tools (And Why You're Overpaying)",{"type":7,"value":8,"toc":227},"minimark",[9,13,24,27,68,86,89,96,101,110,117,120,124,132,135,140,155,158,163,174,179,198,201,204,207,210,213,216],[10,11,12],"p",{},"I keep seeing the same conversation online. \"Have you tried the new Opus?\" \"Gemini 3.5 is unbelievable.\" \"You need the $200\u002Fmo plan or you're leaving money on the table.\"",[10,14,15,16,23],{},"Then I look at the actual usage data from ",[17,18,22],"a",{"href":19,"rel":20},"https:\u002F\u002Fopencode.ai\u002Fdata",[21],"nofollow","OpenCode"," — a terminal-based coding agent platform — and the picture is completely different.",[10,25,26],{},"Market share by model author over the last eight weeks:",[28,29,30,38,44,50,56,62],"ol",{},[31,32,33,37],"li",{},[34,35,36],"strong",{},"DeepSeek"," — 58.4%",[31,39,40,43],{},[34,41,42],{},"Moonshot"," — 24.6%",[31,45,46,49],{},[34,47,48],{},"Qwen"," — 5.9%",[31,51,52,55],{},[34,53,54],{},"Zhipu"," — 5.5%",[31,57,58,61],{},[34,59,60],{},"MiniMax"," — 3.5%",[31,63,64,67],{},[34,65,66],{},"Xiaomi"," — 2.1%",[10,69,70,71,74,75,78,79,82,83],{},"Every single one is Chinese. Every single one is cheap. Average session cost: ",[34,72,73],{},"$0.05."," Token cost per million: ",[34,76,77],{},"$0.14 input, $0.28 output."," Cache ratio: ",[34,80,81],{},"97%."," Tokens per session: ",[34,84,85],{},"4.8 million.",[10,87,88],{},"The models everyone on Twitter is hyping? Barely on the board. Real developers, doing real work, voting with their wallets.",[10,90,91,92,95],{},"This isn't a \"China vs the world\" thing. It's a ",[34,93,94],{},"cost-to-value"," thing. Frontier models are expensive — $2–$15 per session depending on what you're doing. But the vast majority of coding work doesn't need frontier intelligence. Writing a test, refactoring a function, debugging a type error — these are handled perfectly well by models that cost a nickel per session. The only thing holding them back was the tooling.",[97,98,100],"h2",{"id":99},"the-harness-is-the-unlock","The Harness Is the Unlock",[10,102,103,104,109],{},"I've been using ",[17,105,108],{"href":106,"rel":107},"https:\u002F\u002Fpi.dev\u002F",[21],"Pi"," — a minimal terminal harness for coding agents. You bring your own models, tools, and workflows. It supports fifteen-plus providers, so you can switch between a cheap model for daily work and a frontier model for the hard stuff. The harness doesn't care.",[10,111,112,113,116],{},"People use Claude Code and OpenCode for the same reason: ",[34,114,115],{},"they want the agent to actually touch the code."," Not suggest it. Edit it, run it, fix it. That's the real shift — terminal-native agents that operate on your project instead of just chatting about it.",[10,118,119],{},"But the deep truth about why people default to the most expensive tools is simple: it's the path of least resistance. Claude Code ships as a product. Pi is a harness you configure. One is an on-ramp, the other is a workshop. Most people take the on-ramp because it's there, not because they've calculated the cost.",[97,121,123],{"id":122},"the-insane-economics-of-ai-subscriptions","The Insane Economics of AI Subscriptions",[10,125,126,127,131],{},"There's a question nobody wants to ask: why are we spending hundreds or thousands of dollars a month on AI tools to write code for projects that don't even have any users ",[128,129,130],"del",{},"yet","?",[10,133,134],{},"Let's look at the actual numbers.",[10,136,137],{},[34,138,139],{},"Subscriptions:",[141,142,143,146,149,152],"ul",{},[31,144,145],{},"Claude Pro: $20\u002Fmo. Max: $100\u002Fmo.",[31,147,148],{},"ChatGPT Plus: $20\u002Fmo. Pro: $100\u002Fmo.",[31,150,151],{},"GitHub Copilot Pro: $10\u002Fmo. Pro+: $39\u002Fmo. Max: $100\u002Fmo.",[31,153,154],{},"Cursor Pro: $20\u002Fmo. Ultra: $60\u002Fmo.",[10,156,157],{},"Stack two or three of these and you're at $200–$400\u002Fmo easily. Before you've written a single line of code. For a project with five GitHub stars, zero active users, no revenue.",[10,159,160],{},[34,161,162],{},"Frontier API costs:",[141,164,165,168,171],{},[31,166,167],{},"GPT-5.5: $5.00 input \u002F $30.00 output per 1M tokens.",[31,169,170],{},"GPT-5.4: $2.50 input \u002F $15.00 output.",[31,172,173],{},"Claude Opus class models: $5-15 input \u002F $25-75 output per 1M tokens.",[10,175,176],{},[34,177,178],{},"And what OpenCode data shows:",[141,180,181,186,193],{},[31,182,183,184],{},"Average session cost with Chinese models: ",[34,185,73],{},[31,187,188,189,192],{},"Token cost: ",[34,190,191],{},"$0.14 input \u002F $0.28 output"," per 1M tokens.",[31,194,195,196],{},"Tokens per session: ",[34,197,85],{},[10,199,200],{},"You can run a hundred sessions on cheap models for the cost of a single frontier session. And the Chinese models are capturing 100% of real usage on the platform. Not because they're sentimental — because they work.",[10,202,203],{},"And what are you getting for that $100\u002Fmo subscription? A copy-paste workflow from a browser tab. Maybe some agentic features if you use the premium tiers. But the economics make no sense. You're burning capital on inference that could be spent on a dozen cheaper models running in a proper harness.",[10,205,206],{},"Even worse — the more you spend, the less you learn. I see people generating thousands of lines of code they don't understand, shipping features they couldn't explain, building systems they couldn't debug. They're paying a premium to stay incompetent. The expensive model writes the code, the developer approves it blindly, and when something breaks, they have no idea where to start looking.",[10,208,209],{},"You know what forces you to understand your code? Reading it. Running it. Debugging it. A $200\u002Fmo model generating code you skim and accept doesn't make you productive — it makes you a manager of an intern you can't fire who writes code you can't read.",[10,211,212],{},"The people doing this right aren't the ones spending the most. They're the ones who picked a cheap model, configured a harness with real tools, and treat the agent like a junior developer they actually supervise. The cost is incidental. The workflow is the point.",[214,215],"hr",{},[10,217,218,221,222,226],{},[34,219,220],{},"Next up:"," ",[17,223,225],{"href":224},"\u002Fblog\u002Fpi-agent-part-2","Part 2 — How to Actually Use AI Agents (Stop Treating Them Like Magic Wands)"," — why \"Build me a website\" doesn't work, the vibe coder problem, and why supervised collaboration is the real paradigm.",{"title":228,"searchDepth":229,"depth":229,"links":230},"",2,[231,232],{"id":99,"depth":229,"text":100},{"id":122,"depth":229,"text":123},"2026-06-15","OpenCode data shows DeepSeek owns 58% of real usage at $0.05\u002Fsession. The model race is a distraction from what actually matters.","md",false,{},true,1,"\u002Fblog\u002Fpi-agent",{"title":5,"description":234},"AI Agents","blog\u002Fpi-agent",[245,246,247,248],"ai","developer-experience","productivity","software","OpenCode data shows 58% of real coding sessions run on Chinese models at $0.05\u002Fsession. The frontier model race is a distraction — workflow, tooling, and being intentional with your agent matter way more than which model you pick.","o_K81HEL5lm2N0zUA8n6i8aFeYj2Z4zC-h5DBwp9GS8",1784119588408]