GPT-6 ASTRA Habis Berapa Token Buat Jadiin Projek Ini?

Summary

The video tests GPT-6 Astra on a web-development task and estimates the cost of building a small Three.js 3D website using API credits rather than a monthly subscription.

Setup and workflow

• The presenter uses the Nexos AI router through a VPS running OpenCode. The server is a Hostinger KVM2 with 2 CPU cores and 8 GB of RAM.
• After installing Node.js and OpenCode, he connects the model through the provider’s integration settings.
• OpenCode’s Plan mode is read-only and lets the model outline tasks. Build mode can read and write files and execute the project.
• The prompt asks GPT-6 Astra to create and publish a cartoon-style robot-dog website made with Three.js, directly on the VPS. The generated site is served publicly on port 8091.

Cost and token usage

• The test starts with 16.53 credits, described as roughly IDR 300,000.
• The presenter estimates that about 14 credits, or around IDR 200,000, were spent by the end.
• The project used roughly 5 million tokens in total, including input and output, and took about 20–30 minutes to complete.
• The presenter cites an input price of about US$10 per million tokens for GPT-6 Astra, compared with about US$2 per million for GPT-6.1 Soul, which he describes as roughly five times cheaper. The transcript does not give a clear output-token price.
• He contrasts pay-as-you-go API credits with subscription plans, which he says may be more expensive for his use case.

Result and assessment

• The finished site features a detailed 3D robot dog in a room, with a doghouse, smooth camera movement, lighting, shadows, highlights, and animated walking, running, and jumping.
• The robot can interact with a ball using collision detection. The presenter notes some minor floating and observes that the running animation differs from the walking animation.
• He is impressed by the visual quality and says the prompt was relatively simple, though the result is a demonstration rather than a fully production-ready product.
• He assesses GPT-6 Astra as especially capable on this task and says the estimated API cost is low compared with hiring a specialist Three.js freelancer. He suggests the project could be expanded with more features.

Guide and tutorial points

• The video demonstrates installing Node.js and OpenCode on a VPS, connecting a model through an AI router, and choosing between Plan and Build modes.
• It also walks through a VPS purchase flow and mentions KVM2 as a suitable option for experimentation. KVM4 is suggested for users who want more memory to run a local LLM.

Main speaker and source: Dea Afrizal, the video’s presenter and narrator.