Artificial Analysis finds Chinese AI models heavily undercut US rivals, with API rates starting at $0.10 per million tokens

@zerohedge When you realize China’s AI charges cents while your API bill looks like a mortgage payment 💸🤖😂

@zerohedge Don’t use this chart to compare cost, use metrics like this

API prices of key AI models: US vs China

Assuming each token has the same value is a false comparison. One model could be infinitely better than another at solving a set of problems crucial to its customers. So this type of token-only price comparison is really apples and oranges. Watch what you eat, too much fruitcake can be lethal to one’s longterm health.

@zerohedge Price per token is not a useful comparison without knowing token usage.

Thank you for coming to my TED talk.

@zerohedge GPT 5.6 on Thurs June 25, gives Fable 5 a max of Weds June24. Before GLM 5.2 eats all their lunch.

@zerohedge They're public too.

http://2513.HK http://100.HK BABA http://0700.HK

@zerohedge GLM 5.2 is fantastic, I'm cancelling my Claude subscription.

@zerohedge Frontier for frontier problems and use alternative for non-frontier problems. The US needs to find a domestic similarity to chinas lagging but cheaper model.

@zerohedge Woof

@zerohedge Details of my stock holdings are as follows

⬇️

@zerohedge Optimizing token usage is becoming an important skill in the post token-subsidy era. Use the right model for the right job.

@zerohedge Price will dictate the winner unless intelligent level is so high to the point where the prices make sense

@zerohedge Only reason China seems to have better AI is because they don’t have us laws to worry about when stealing data.

@zerohedge MiniMax is so cheap for being an Opus level model based on numbers from that table, at only 5% the cost.

@zerohedge And Trump keep crying US is leader in ai 😭😭😭

@zerohedge Yeah, well, you get what you pay for right now. Right now that gap is what electric cars were 10 years ago. Open weight models aren't at frontier level, but may well be there next year.

@zerohedge 👀👀

Anthropic

From FundaAI (big X follow amongst tech pros)



If I’m reading this correctly, the second derivative growth has slowed materially. Agree the 47bb figure was from May 1st I am told. So 47>53 is 12% growth. April was 30 >47 which was 56% growth. Massive slowdown. I’m guessing the big bump was from all of the agentic “first timers” trying it out and then slowing down materially. My guess is the next print will be a negative number.

They claim their ests have been extremely accurate