Continuing from yesterday, after failing to get a Coding Plan yesterday, I turned on my computer early again today to wait for a chance to grab a spot. I took a look, and the most sought-after ones were Zhipu and Alibaba Cloud Bailian. You don't need to fight for the same thing on Volcano Engine Cloud.
As the saying goes, everything has an impossible trinity. For Coding Plans, that impossible trinity is: good, cheap, and no need to fight for a spot. Alibaba Cloud started its allocation at 9:30, while Zhipu started at 10:00, so the times were perfectly staggered.


In the end, both companies toyed with us. Alibaba Cloud Bailian simply stopped selling the Lite plan, leaving only the Pro plan at 200 yuan per month. As for Zhipu, the webpage became inaccessible the moment it hit 10:00. Refreshing constantly was useless; I couldn't get in until around 10:10, by which time it had long been sold out. I have no idea which script gurus managed to grab one.
With no other choice, I had to settle for MiniMax. Recently, I had often seen posts about MiniMax claiming that the MiniMax-M2.7 model had caught up with the quality of the first tier, so I wanted to see just how impressive it really was.

I spent 26 yuan on the cheapest plan and, following the official documentation, configured the key and model in Claude Code. Eager to get started, I began writing business code to optimize the Auth for the current project.
After running for a few minutes, the result came out. All I can say is that I was terribly disappointed. I carefully reviewed the code generated by MiniMax-M2.7, and in many places it was even more foolish than what I would have written myself. It was so absurd that I could only laugh in exasperation.
In places where ESLint reported errors, some were merely very basic type warnings. It didn't bother to think twice and simply appended as type declarations, without considering or searching the context to see whether that type had already been declared somewhere.
The more I thought about it, the angrier I got. On top of that, MiniMax's Token Plan doesn't support refunds. It felt as though I had spent more than twenty yuan hiring an extremely stupid employee and then still had to pay him a salary for his probation period.
Going from full of anticipation to deeply disappointed, all I can say is that domestic large language models still have a very long way to go. Even Xiaomi's newly released MiMo-V2-Pro yesterday was underwhelming. Sure enough, when it comes to productivity tools, you still have to go with the best.
Claude Opus 4.6 remains an invincible presence in the Coding field. The only one that can barely compare with it is GPT 5.4. It doesn't matter how far ahead other models are in benchmark scores; once they are put into real-world business scenarios, it becomes immediately obvious who is swimming naked.
That secondary Github Copilot account still hasn't been banned today, so I can temporarily continue using the Opus model. I've already made up my mind: even if the account gets banned after a while, I will still choose Copilot through the official channel rather than opting for another domestic large language model.
The A-share market officially fell below 4,000 points today, dipping as low as 3,955. It fell all the way down, and I saw no funds supporting the market in the afternoon. From a technical perspective, it has now broken below all the candlesticks today, so it will probably need to undergo an adjustment for quite some time.
The two oil giants did not surge against the trend today. Instead, they led the plunge. Yangtze Power held up for half the day but eventually turned negative, ending the session with a loss of more than 800 yuan. Since March began, this was the first time I had seen such a large retracement in profits, and I was quite unaccustomed to it.
Today I also took a look at the Hong Kong stock market through MiniMax and found it rather interesting. A company like MiniMax actually had its share price speculated up to more than 1,000 yuan, making it the company with the highest share price in Hong Kong stocks. Zhipu is currently ranked third.
All I can say is that today's capital markets rely entirely on telling tall tales. Domestic large language models have never lost when it comes to benchmark scores, but in actual use, they fall silent every single time. If only they would spend their marketing budgets on research and development, perhaps their products wouldn't be like this.
Xiaomi is the same. It released MiMo-V2-Pro yesterday and a new version of the Xiaomi Su7 today, yet the capital market showed no confidence in it at all. Its share price plunged nearly nine points today, with the entire weekly chart breaking down; from last year's high, it has almost been cut in half.
I still remember that I used to be a Mi fan. No matter what electronic product I wanted to buy, if Xiaomi offered one, I would basically choose Xiaomi. Even now, I still buy quite a lot of products from the Mi Home line, but I have absolutely no favorable feelings toward Xiaomi's current marketing.
Ever since last year's controversy over the tiny text, I have felt disgusted by these seemingly pathological marketing tactics. When will domestic brands truly learn to keep their feet on the ground instead of always promoting this so-called overwhelming lead? Only by facing their own shortcomings can they set out better.