AI’s “premium pricing” just took another hit
Saturday, October 17: DeepSeek upgrades to V4.1-Flash-1017
Just 5 weeks since the previous version, with no API price increase
The numbers are solid
77.0% on the DeepSWE coding benchmark, up from 74.2% in the previous version
Beats Claude Opus 5’s 74.0%
Cost per task: 3.3 yuan → 1.3 yuan, down 60%
Speed: 237 tokens/second; weights open-sourced under MIT
The timing matters even more
This week, US AI stocks were just hammered
On Friday, the Nasdaq fell 2.14% and the Philadelphia Semiconductor Index dropped 3.8%
CoreWeave fell 25% in a week, while bond yields surged to 15%
The market is starting to doubt whether AI can actually make money
Then this Hangzhou-based company says: capabilities are improving, and costs can still come down
But the other side is the real point
As tokens get cheaper, compute providers have to sell more to maintain revenue
When efficiency outpaces demand, GPU prices and volumes come under pressure
That’s why US stocks may revalue Nvidia on Monday
My view
Total compute capacity won’t shrink; an explosion in applications will make up the difference
But the era of outsized profits at the model layer is over—intelligence is becoming a utility, like water, electricity, and gas
From now on, the winners will be those who turn intelligence into products
This directly affects crypto, too
AI-themed coins will track sentiment around US AI stocks in the short term
The real price discovery comes when US markets open Monday night
Don’t chase the news on Saturday; wait for confirmation on Monday
Goldman says AI stocks have seen their most extreme sell-off since ChatGPT
I partly agree—sentiment-driven sell-offs often create opportunities
But let prices find their own level; don’t try to call the bottom for the market
Intelligence is getting cheaper
Is this deflation or wider adoption? 😂
$TAO
#DeepSeek发布更新版AI模型 #AI #英伟达
Saturday, October 17: DeepSeek upgrades to V4.1-Flash-1017
Just 5 weeks since the previous version, with no API price increase
The numbers are solid
77.0% on the DeepSWE coding benchmark, up from 74.2% in the previous version
Beats Claude Opus 5’s 74.0%
Cost per task: 3.3 yuan → 1.3 yuan, down 60%
Speed: 237 tokens/second; weights open-sourced under MIT
The timing matters even more
This week, US AI stocks were just hammered
On Friday, the Nasdaq fell 2.14% and the Philadelphia Semiconductor Index dropped 3.8%
CoreWeave fell 25% in a week, while bond yields surged to 15%
The market is starting to doubt whether AI can actually make money
Then this Hangzhou-based company says: capabilities are improving, and costs can still come down
But the other side is the real point
As tokens get cheaper, compute providers have to sell more to maintain revenue
When efficiency outpaces demand, GPU prices and volumes come under pressure
That’s why US stocks may revalue Nvidia on Monday
My view
Total compute capacity won’t shrink; an explosion in applications will make up the difference
But the era of outsized profits at the model layer is over—intelligence is becoming a utility, like water, electricity, and gas
From now on, the winners will be those who turn intelligence into products
This directly affects crypto, too
AI-themed coins will track sentiment around US AI stocks in the short term
The real price discovery comes when US markets open Monday night
Don’t chase the news on Saturday; wait for confirmation on Monday
Goldman says AI stocks have seen their most extreme sell-off since ChatGPT
I partly agree—sentiment-driven sell-offs often create opportunities
But let prices find their own level; don’t try to call the bottom for the market
Intelligence is getting cheaper
Is this deflation or wider adoption? 😂
$TAO
#DeepSeek发布更新版AI模型 #AI #英伟达