☕️🎃🐾 "The Rainy Night Café"
It was warm and cozy inside the café. Snuggled into the large sofa and cradling a steaming cup of coffee, I said happily below about opus5.5, grok bot, and codex astra.
As mentioned yesterday, ChatGPT has introduced the new GPT6-Sol and GPT6-Luna models; it looks like the mid-tier Terra models will eventually be phased out.
I found that running the same program on grok4.7 took longer than on grok4.6 without offering any significant improvement; I'll try the image and video tests later. I heard today that SOL and Opus 5.x are also going to be released.
behind Jev truly emerges when developers begin to consider which processes should be codified into programs, which decisions still require semantic reasoning capabilities, and how to ensure the entire system can be repeatedly executed, tested, and monitored.
I haven't really looked at X properly in a long time, and I rarely come online. I only come here to read articles and then go back to work. Looking back on the past year, I've made many new attempts and gained a lot, but ultimately it's just a process.
Some get lost in AI, thinking "making games is so easy," simply copying games from GitHub and those they've played. What makes them think someone with less than 10 years of exp can do better level design, sound effects, and art than them, especially since I am good in programming
Status.claude.com you will find out the models that degraded these days, no wonder my work quality becomes so terrible.
After I did tons of test, anthropic benchmark lies, opus5 is not as good as fable5, the simple question you ask urself, if opus5 better than fable5, anthropic should take down the price in fable5. It’s just let me felt I bought iPhone 17 pro max, and they release iPhone 17 e..
Did tons of test, those benchmark lies, opus5 is way too stubid, even gave it same spec, same request, even asked opus5 to read previous fable5 high effort made to be opus5 example, opus5 xhigh still made same mistake just like opus 4.8 made. Totally wasted my time and token
模型 生產穩定度 主要問題
Claude Sonnet 4.6日常 90% 任務夠用，指令遵從最穩
Claude Opus 4.6 重活 1M ctx TerminalBench 第一
o4-mini** CP 值王 中等複雜度、高吞吐
Gemini 2.5 Pro便宜可用 有免費 tierSWE-bench 78%
Claude Opus 4.8需釘版本 | Grader awareness + 延遲
what I'm doing？I'm struggling with the 5-hour limit reset, switching models to work on projects. Claude Fable5 was delayed another week, and the Codex ChatGPT 5.6 solution was very generous, so I quickly took advantage of the time to finish the projects I wanted to work on.
如何才能發揮 Pro 訂閱的 Anthropic Claude、Fable5 和 Opus 的合作優勢？首先，繪製樹狀圖明確目標，建立主題。token 很昂貴，訂閱pro, 要有plan, 有程式基礎，知道如何完善Claude和memory.md建立後開工，大型程序拆解成小塊,將來讓模型讀取改動部分可以省掉讀取不必要的程序碼問題 👇