<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>DeepSeek on Hummingbird Labs</title><link>https://hummingbirdlabs.github.io/tags/deepseek/</link><description>Recent content in DeepSeek on Hummingbird Labs</description><generator>Hugo</generator><language>en-us</language><lastBuildDate>Sun, 24 May 2026 09:36:00 +0800</lastBuildDate><atom:link href="https://hummingbirdlabs.github.io/tags/deepseek/index.xml" rel="self" type="application/rss+xml"/><item><title>Deepseek V4 Pro Price Drop Again on May 23, 2026</title><link>https://hummingbirdlabs.github.io/posts/post-006/</link><pubDate>Sun, 24 May 2026 09:36:00 +0800</pubDate><guid>https://hummingbirdlabs.github.io/posts/post-006/</guid><description>Deepseek V4 Pro pricing reduced to 25% of original price. Exploring AI model cost trends, WPF development challenges with LLMs, and why detailed debug logging is essential for AI-assisted coding.</description></item><item><title>More on TRAE China Version: Free Models Are Great But Slow</title><link>https://hummingbirdlabs.github.io/posts/post-005/</link><pubDate>Fri, 22 May 2026 21:16:00 +0800</pubDate><guid>https://hummingbirdlabs.github.io/posts/post-005/</guid><description>A detailed performance comparison of TRAE China&amp;#39;s free LLMs vs paid models: DeepSeek V4 Pro with think mode takes 50 minutes for large tasks, while free Qwen3.6 Plus is significantly faster with shorter queue times. Real data from actual coding projects, plus thoughts on DeepSeek&amp;#39;s mission to democratize AI.</description></item><item><title>Why I Use TRAE: Free LLMs, Stability, and 1M Token Context</title><link>https://hummingbirdlabs.github.io/posts/post-004/</link><pubDate>Fri, 22 May 2026 14:37:00 +0800</pubDate><guid>https://hummingbirdlabs.github.io/posts/post-004/</guid><description>A hands-on review of TRAE IDE&amp;#39;s China version: 14 free LLMs including DeepSeek v4 Pro and Qwen 3.6 Plus, 3-minute average wait times for free tier, fewer task freezes vs GitHub Copilot, and how DeepSeek&amp;#39;s 1M-token context window beats Claude&amp;#39;s 168K limit for large codebases.</description></item><item><title>Using Qwen 3.6 Plus: Great but a Bit Expensive</title><link>https://hummingbirdlabs.github.io/posts/post-003/</link><pubDate>Fri, 22 May 2026 08:22:00 +0800</pubDate><guid>https://hummingbirdlabs.github.io/posts/post-003/</guid><description>Hands-on cost analysis of Qwen 3.6 Plus for large-scale C# development: why 876K tokens for 1.7 RMB looks cheap but 30 RMB per engineering task adds up fast, and how a dual-model strategy with DeepSeek cuts the bill.</description></item><item><title>DeepSeek v4 Pro, Qwen 3.6 Plus, or Others: Which Should I Use?</title><link>https://hummingbirdlabs.github.io/posts/post-002/</link><pubDate>Thu, 21 May 2026 19:11:00 +0800</pubDate><guid>https://hummingbirdlabs.github.io/posts/post-002/</guid><description>Real daily usage data comparing DeepSeek v4 Pro vs Qwen 3.6 Plus for AI-assisted coding: token counts, RMB costs, and which model gives better value after GitHub Copilot&amp;#39;s June 2026 pricing shift.</description></item></channel></rss>