Concrete Data Proving AI's Decline in Intelligence: Claude Opus 4.5 This data shows the Marginlab team using Claude Code's Opus 4.5 daily to perform pass/fail tests on 50 questions in the SWE-Bench-Pro test. The data shows that the pass/fail rate has dropped from 60% in early January to 54% currently. The rate of AI decline is 10%.

Here are a few recent observations: 1. For coding ability, CodeX GPT 5.2-CodeX-High is the strongest. It's suitable for writing complex products or finding hard-to-find bugs. It can solve several bugs that others couldn't. The downside is that it's extremely slow; don't use it for minor modifications. Modifying a small part can take a week to read a novel like "A Record of a Mortal's Journey to Immortality." 2. For coding ability, Claude Opus 4.5 is the next best. Its quality and capabilities are second only to the top, but it's much faster. It can also be used in antigravity mode. 3. Gemini 3 Pro. It's the strongest for front-end development. I used it for all my front-end work, and its aesthetics were generally good. Besides coding, it's also highly recommended for other scenarios, such as daily conversations and connecting to the Google ecosystem. Google is incredibly generous; you can use it for free for a year with a US education qualification. 4. Claude Sonnet 4.5 is rather awkward. I haven't used it for a while. Its capabilities aren't as strong as the previous three. Using the API is quite expensive. For a monthly subscription, Opus 4.5 is recommended. 5. A cost-effective option is GLM 4.7. Their coding capabilities are decent, and their models are updated quickly. Most importantly, it's affordable, less than 20 RMB per month. You'll likely never use it all. It can be used for minor modifications, etc. 6. Don't trust AI rankings or those self-media outlets too much. They claim every new AI is the best in the world. I don't know if they've been paid to do so or if they genuinely don't understand. They're just wasting other people's time and energy.

Found a great crypto blogger, the title of Big Cousin's book. Don't miss my crypto life if you missed Binance Life. A new crypto life in 2026, still at the bottom! Worth buying in! Recently, older cryptocurrencies have been experiencing a resurgence, I believe 2026 will also be a banner year for crypto. My crypto life = starting a 100x life! #meme $BinanceLife {web3_wallet_create}(560x5b68e58418f0cf6111922ea8ff0b5e88e8894444)

The most frustrating thing for Vibe coders is that every few days later, another model emerges that dominates the leaderboards, boasting world-leading metrics. You can't possibly try every single one, so you rely on social media bloggers. You'll find that most of them are just hype. Then, after finally settling on a suitable model and product, you're ready to get down to business. But within days, the model seems to have lost its intelligence. It's truly speechless. Recently, I discovered that Gemini3 Pro and Claude 4.5 have both become less intelligent. Yesterday, I used Gemini3 with antigravity to write a task, and it kept encountering problems. Okay, I tested a few commonly used ones, and the results were: 1. Gemini3 has indeed lost its intelligence. 2. I tested Claude Code using Claude 4.5, and it also showed severe intelligence loss. 3. I used GLM 4.7 in Claude Code, which recently appeared claiming to be the world's best. While Claude 4.5 found a few more issues, it didn't find 1/3 of them. Among the four cursors, GPT 5.2 performed the best, missing only one minor issue; the others were all found. Based on this, GPT 5.2 is the most reliable. The only headache is that GPT 5.2 code is too concise; much of it is incomprehensible, making it a headache to read. 😀

A group of high-ranking Yapper accounts on Kaito have been discovered, exhibiting very similar behavior: Followers: Approximately 20,000 to 50,000 Tweet views: Around 1,000 per post Many comments under each tweet Also, they tweet an incredible amount every day, about 20 tweets a day! Truly workaholics!

As expected, Google's AI team knew the inside information first. Just this December, Google revealed the development of a new AI browser. OpenAI also has Atlas. It's predictable that after AI coding, AI browsers will become the next battleground for tech giants.

The cryptocurrency market needs a phoenix-like rebirth but it must first endure immense pain.

There's a very simple yet highly effective method for predicting BTC's future trend: It's by comparing it to the US stock market. I've clearly marked this in the chart. The red boxes indicate potential turning points in the US stock market. The bottom line represents the US stock market, and the blue line represents BTC. For example, around October 2023, the US stock market fell for 10 consecutive days, while BTC rose steadily, experiencing a significant surge. The same pattern occurred in May 2025. Conversely, the same applies to BTC's corrections. Of the three red lines, the green line represents the US stock market's upward trend, while BTC is falling, followed by another significant drop. The recent decline is also very evident; the US stock market has been rising for many days, while BTC has been trending downwards.

The Gemini 3 has finally arrived! I used it this morning to resolve a few issues, and it's great. However, what I don't understand is that Claude Sonnet 4.5 seems to have deliberately lowered its intelligence to celebrate the Gemini 3 release. I'm speechless.

If you want to buy a coin and try to buy it when the price drops slightly, but it always surges before reaching your target price, it indicates a bull market. If you want to sell a coin and try to sell it when the price rebounds slightly, but it always starts to fall before reaching your target price, it indicates a bear market. If you frequently encounter either of these situations, sorry, my friend. It means you're a big loser.