https://blog.jqknono.com/tags/short-post/Recent content in Short Post on jqknono BlogsHugo -- gohugo.ioen-ushttps://blog.jqknono.com (jqknono)https://blog.jqknono.com (jqknono)Wed, 01 Jul 2026 00:03:08 +0800https://blog.jqknono.com/blog/2026/07/01/xhs-6a43e8bc00000000200388fa/Wed, 01 Jul 2026 00:03:08 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/07/01/xhs-6a43e8bc00000000200388fa/What matters is the model; the gap between harnesses isn’t that big. Recently A/ has opened up some reverse-proxy access, so it’s worth trying other harness tools. opencode, copilot cli, etc. are all good.https://blog.jqknono.com/blog/2026/06/26/xhs-6a3df619000000002003acc1/Fri, 26 Jun 2026 11:46:33 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/06/26/xhs-6a3df619000000002003acc1/I used to most recommend chatgpt plus, but the other day chatgpt banned my codex. Logging into codex again requires SMS two-factor authentication. Since it was a number from a code-receiving platform, openai does not support self-service phone number rebinding…https://blog.jqknono.com/blog/2026/06/17/xhs-6a32883d000000002003a02c/Wed, 17 Jun 2026 19:42:53 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/06/17/xhs-6a32883d000000002003a02c/The second screenshot is enough to prove it’s genuine, the familiar 1302 flavor. Stable to use on 6.18, speed is quite good, maybe the supplier no longer uses z.ai. In practice, it’s stronger than deepseekv4pro in coding, script error rate…https://blog.jqknono.com/blog/2026/06/14/xhs-6a2ec48e000000000603399f/Sun, 14 Jun 2026 23:11:10 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/06/14/xhs-6a2ec48e000000000603399f/Requires configuring scheduled tasks on a computer; you can look into it yourselfhttps://blog.jqknono.com/blog/2026/06/09/xhs-6a2786ef0000000008001fe8/Tue, 09 Jun 2026 11:22:23 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/06/09/xhs-6a2786ef0000000008001fe8/Not cheap, especially the way token consumption is counted—both input and output are counted within the plan. I often use 20 million tokens in a single request, and the 59 yuan tier only covers about 3 of my requests. For coding, roughly 99% of input is input cache,…https://blog.jqknono.com/blog/2026/05/22/xhs-6a106d2d0000000007024a1e/Fri, 22 May 2026 22:50:21 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/05/22/xhs-6a106d2d0000000007024a1e/2.5折 directly becomes 1/4 permanently, no need to rush purchase, no packages, no need to grind 5h daily or weekly quota, no need to calculate off-peak/peak multipliers, no need to calculate token-to-credit exchange rates, direct retail equals wholesale, no tricks, all sincerity…https://blog.jqknono.com/blog/2026/05/08/xhs-69fd5f470000000022027248/Fri, 08 May 2026 11:57:59 +0800https://blog.jqknono.com (jqknono)https://blog.jqknono.com/blog/2026/05/08/xhs-69fd5f470000000022027248/My main use case is writing code, and my projects are mainly command-line tools. Tencent HY3 Preview performs very well in my use case, with good instruction following and dialectical logic. Baidu’s CoBuddy recently launched on openrouter…