@evan-dong

DeepSeek-V4 Deep Dive: Huawei Ascend Optimization, Sub-$0.30 Pricing, and the Architecture That Makes It Work

5h ago · 14 min read · DeepSeek released V4 on April 24, 2026. If you only read the headline numbers — 1M token context, Agent-tier reasoning, $0.28/M output tokens on Flash — you'd think this is just another price war move. It's not. V4 is the first Tier-1 model built fro...

GPT Image 2 + Seedance 2.0: The Two-Stage AI Video Production Workflow

1d ago · 3 min read · GPT Image 2 + Seedance 2.0: The Two-Stage AI Video Production Workflow Most AI video workflows fail at the same point: they ask one model to solve character design, composition, motion, editing rhythm, and sound simultaneously. The uncertainty compou...

Google Deep Research: Agentic Research Platform, Not Just a Better Summarizer

2d ago · 3 min read · Google Deep Research: Agentic Research Platform, Not Just a Better Summarizer Google's latest Deep Research upgrade — Gemini 3.1 Pro backbone, Deep Research Max tier, MCP support, multimodal grounding, private data integration — is being positioned a...

Claude Design: LUI Over GUI — What Anthropic's Design Tool Actually Changes

4d ago · 3 min read · Claude Design: LUI Over GUI — What Anthropic's Design Tool Actually Changes Anthropic launched Claude Design this week. Within hours, Adobe and Figma shares were moving. The market reaction probably overestimates the immediate product threat. But it ...

GPT Image 2: Why Text Rendering in AI Images Is the Real Breakthrough

5d ago · 3 min read · GPT Image 2: Why Text Rendering in AI Images Is the Real Breakthrough OpenAI hasn't officially announced GPT Image 2, but the community has been cataloging outputs from ChatGPT A/B tests and API metadata that show something meaningfully different: te...