用户实测 Gemini 4 Argon：无搜索单次推理答冷门知识

kimse /\ nobody @Misteriazq原推发布于 2026.10.19

在 X 上打开原推

Gemini 4 argon is at another level. She knows almost everything, no it is not a joke literally it even knows most obscure things on the world. From deep nuances of esoteric traditions to most specifical biological mechanisms which needs really deep research on these topics but gemini 4 argon literally going all correct without any mistake without any **search** tool in a single forward pass in 3 seconds. It could even say what i was working on 2 days ago with just one mathematical pattern i observed no info about the source of pattern, only info was that i was interested in physics too. Btw it didnt say you worked on that it said if you didnt look at it fits these things which were my simulation subjects from yesterday when it said. If we come coding part it is amazing on coding too. Literally it can do things which opus 5.5 struggles or proposes more basic but inferior things. Another thing it is literally so good at orchestrating other agents like claude and gpt with nested subagent flows. It can think creating daemon for managing lots of agents at once coherently without needing to say it specifically. On transparency part it is the thing gemini mogs every other model it cna explain what it did and why it did coherently without lacking important details or choice of abstraction. But it is way more better than that because it can explain anything both intuitively and technically enough which opus 5.5 trades one with other. On tone or pair coding part it just feels good, too good. It doesnt do classical llm quirk which is when you said it is x and it says it isnt y{¬x} it is x quirk i mean. It doesnt do that and while preserving correctness, this is literally sota on communication. And even more importantly it is not doing that quirk even in chain of thought, This is more important than you can actually think becuase it isnt eliminating possibilities or methods for only sake of having negative attention space boost which could cause overlooking a good method or being too rigid on standart technique, this quirk caused opus 5.5 missing lots of algoritmic opportunity and gemini 4 argon is simply just not doing that thing so it doesnt overlook opportunities. Another part this is so important, self esteem it is well calibrated in gemini 4 argon. For context i am tlaking about the thing gpt wont believe any idea can work in real world it just wont believe it. It believes everything tend to fail except already succeed things. Opus 5.5 just going neutral on that which is like "i dont know but can, we need to try first". Now with argon it is just chill but not killing excitement just says yes it can if we do these things correctly there is no reason for failure except random things. This is awesome i love it. Congratulations on gemini 4. It is not just a good model sir, it is the best model i have ever seen. @DynamicWebPaige @rodydavis @OfficialLoganK @zacharynado