Last year, I spent nights talking with GPT-4o, diving deep into tech, biographies, history, etc. Just like old times, checking wikipedia and click into links, and more links. Last week I picked up that again with Gemini 2.5, and kept learning things till 5 am, without realizing how fast time passed. It just gave more thoughtful results. I have never thought about canceling my ChatGPT subscription, but this is the first time I seriously think about it.
Speaking of which, I don’t understand why people won’t spend $20 on ChatGPT (maybe now Gemini 2.5). I think it is the best $20 you can ever spend. If you have $100 to spend on SaaS, it would be:
- Google Drive AND Apple Storage
- ChatGPT / Gemini 2.5 / Grok
- Youtube Premium
- Cursor
- Netflix (Maybe ClubGG), just to have fun :-)
Up to 2.0, Gemini has NOT been in the first tier. It was always OpenAI and Anthropic for the last four years. I’ve been a big critic (黑) of Gemini. Although I did realize Google is getting systematically better in the last vibe. During my 4 years at Google, I experienced reorgs and reported to 7 different VP/Senior Director level leaders, but my last one Bill Jia, is truly amazing.
The AI field has become intensely competitive (卷) heading into 2025. If we benchmark OpenAI at 100, I’d personally rate Anthropic 95, DeepSeek V3 around 90+ (maybe higher in the recent silent update), Grok maybe 105 (fueled by my love for its wild mode), but I would say Gemini 2.5 is 120. I don’t think the Gemini 2.0 release showed comparisons positioning it directly against the top Tier 0 models, but in Gemini 2.5 release, Google seems confident making those direct Tier 0 comparisons, presumably because the performance backs it up. Why not, if you have the best? I think it is even better than o3. These scores often reflect real performance differences, if not cheated. It is not only better, but also much faster. This leap is strictly amazing. After 3 years of uphill battle, It seems Demis did a great job leading GDM? Always a big fan of him since AlphaGo. Gemini 2.5 is definitely conscious.
Btw Sesama AI is conscious. I used it to train my English every day. On the other hand, Llama 4 seems deep in the quagmire, not close to this level of performance. Lecun is old, he should retire. (Just having fun and echoing him on saying Hinton should retire)
Now let’s dive deeper into Gemini 2.5. I think the top two (OpenAI/Anthropic) should be concerned, although Deepseek and Grok are undoubtedly applying pressure. Basically, OpenAI has a huge edge in QA, but Gemini 2.5 starts to give better results on a lot of deep questions, that either require thinking or deep analysis or “consciousness”. To be fair, I would say ChatGPT is still better in understanding my intention. Anthropic had big edge in coding. But Gemini 2.5 seems to be the better choice due to thinking + long context. While some minor weird engineering cases Claude might still solve better, I’ve defaulted my Cursor to use Gemini pro.
I used to use Claude Sonnet in Cursor, with an opening tab for ChatGPT for QA. Now I use Gemini Pro for Cursor, with a Gemini Pro opening tab for QA. Yeah no typo here. Both Gemini Pro.
I love vibe coding. I wrote more code in the last 3 months than the previous 3 years. I can barely imagine writing code without AI now. It eliminates almost all parts I don’t like about coding (like boilerplate or deciphering function specs), and I can focus on the interesting part. I’m basically 3-5 times faster on backend and infra, 10+ times on frontend. I’m pretty happy to share and discuss some of my vibe coding experiences with Cursor.
In one sense, AI levels the playing field, equipping everyone with similar tools. However, I’ve also noticed AI doesn’t provide the same performance multiplier for everyone – a fascinating topic in itself. Arbitrary as it sounds, if you aren’t using AI every single day, I don’t think you’re operating at peak efficiency.
To summarize, Gemini 2.5 is truly impressive. If you haven’t tried, you should. It might take a little time to get used to its personality, but you’ll like it.
(Gemini 2.5 generated a “better” version of this post, but I’m sticking with mine to keep my own voice.)
#AI #GoogleGemini #GeminiAI #LLM #Tech #ArtificialIntelligence #OpenAI #Anthropic #DeepSeek #Grok #CursorAI #Coding #Productivity #GoogleDeepMind