DeepSeek V4: SOTA Coding Agent at 12x Lower Cost

DeepSeek V4 Pro matches GPT-5.5 and Opus 4.7 on agentic coding benchmarks at 3x lower cost. In this video, we compare Terminal-Bench, SWE Verified, and MCPAtlas scores across all SOTA models, analyze pricing, and demo DeepSeek V4 with the Pi coding harness. --- 🤓 *Topics Covered* - DeepSeek V4 benchmark comparison vs GPT-5.5 and Opus 4.7 - Terminal-Bench, SWE-Bench Pro, and agentic benchmarks explained - Pricing analysis: $1.74/M input vs $5/M for GPT-5.5 - Setting up DeepSeek V4 with Pi agent harness via Hugging Face --- ⏱️ *Timestamps* 0:00 Introduction 0:01:15 What is DeepSeek 0:03:12 Benchmarks 0:09:08 Pricing 0:10:23 Pi coding harness demo 0:15:30 Test results and outro ## Connect with me X: https://x.com/_alejandroao LinkedIn: https://www.linkedin.com/in/alejandro-ao/