#opus4 — Public Fediverse posts

1M context is now generally available for Opus 4.6 and Sonnet 4.6

https://claude.com/blog/1m-context-ga

#HackerNews #1MContext #Opus4.6 #Sonnet4.6 #TechNews #AIUpdates

🎩🔬 Ah, the esteemed #Claude - savior of #science and destroyer of data integrity, now with new and improved "better-than-human" skills! Scientists are absolutely thrilled to have #AI interpret their figures because, let's face it, who needs human brains when you have Opus 4.5? 😂🔍
https://www.anthropic.com/news/accelerating-scientific-research #DataIntegrity #BetterThanHuman #Opus4.5 #HackerNews #ngated

Opus 4.5 is going to change everything

https://burkeholland.github.io/posts/opus-4-5-change-everything/

#HackerNews #Opus4.5 #ChangeEverything #Innovation #TechNews #FutureTrends

Measuring AI Ability to Complete Long Tasks: Opus 4.5 has 50% horizon of 4h49M

https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

#HackerNews #MeasuringAI #LongTasks #Opus4.5 #AIResearch #TaskCompletion

https://blog.gslin.org/archives/2025/11/23/12746/%e6%b8%ac%e8%a9%a6%e5%90%84%e5%ae%b6-llm-%e5%b0%8d%e3%80%8ccalifornium%e3%80%8d%e7%9a%84%e7%bf%bb%e8%ad%af/

測試各家 LLM 對「Californium」的翻譯

#ai #anthropic #californium #claude #gemini #Gemini2.5Flash #Gemini2.5Pro #Gemini3ProPreview #google #gpt #Gpt4.1 #Gpt5 #Gpt5.1 #llm #mistral #model #openai #opus #Opus4.1 #sonnet #Sonnet4.5

After a few weeks using #Claude4 and #Opus4 I found the new models are overwhelming and unnecessarily verbose. Well constrained and explained tasks become very long codebase including unspecified scenarios handling, simple internal procedures becomes unnecessary function/snippets and out-of-scope unit testing are included. I understand that the great #LLM challenge is to provide information that users poorly request but I think this is a bit too much for a reasonable code helper tool. ⛔