41–50 of 61 posts
Earlier quoted context omitted.
For reference (input cache hit, input cache miss, output): Deepseek V4 Flash: $0.0028, $0.14, $0.28 Deepseek V4 Pro: $0.145, $1.74, $3.48 GPT 5.5: $0.5, $5, 430 GPT 5.5 Pro: $0.5, $30, $180 Claude Sonnet 4.6: $0.30, $3, $15 Claude 4.7 Opus: $0.5, $5, $25
GPT 5.5: 430 or $30?
Earlier quoted context omitted.
ChatGPT (the setting is shared with Codex) and Claude (shared with Claude Code) also have sharing enabled by default, so why aren't they cheaper?
There's evidence various third-party models (including Deepseek) used distilling in training, based on models from those leading services. So they have more flexibility with pricing.
Earlier quoted context omitted.
Energy is like 10-20% of the cost of AI. The rest is mostly hardware depreciation.
So you think they’re running the same types of state of the art Nvidia deployments?
Nvidia H100: Typically priced around $25,000–$30,000 (global MSRP).
Huawei Ascend 910C: Reported to cost roughly $28,000, yet it delivers only 60% of the inference performance of the Nvidia H100.
Google's TPUs are significantly cheaper for Google for inference. That's pretty much it.
There's a reason nVidia has an 80% margin right now.
MiMo-V2.5 Series Input (Cache Hit) Input (Cache Miss) Output mimo-v2.5-pro $0.0036 $0.435 $0.87 mimo-v2.5 $0.0028 $0.14 $0.28
For reference (input cache hit, input cache miss, output): Deepseek V4 Flash: $0.0028, $0.14, $0.28 Deepseek V4 Pro: $0.145, $1.74, $3.48 GPT 5.5: $0.5, $5, 430 GPT 5.5 Pro: $0.5, $30, $180 Claude Sonnet 4.6: $0.30, $3, $15 Claude 4.7 Opus: $0.5, $5, $25
Earlier quoted context omitted.
This runs counter to the last 50 years of American propaganda espousing the inefficiency of government. If the Chinese government can just throw money at industries and have them flourish, why can't other governments?
Any government can and does regularly throw money at industries to make them flourish. The American propaganda claims that this is less efficient than letting market forces decide which companies win.
Earlier quoted context omitted.
ChatGPT (the setting is shared with Codex) and Claude (shared with Claude Code) also have sharing enabled by default, so why aren't they cheaper?
There's evidence various third-party models (including Deepseek) used distilling in training, based on models from those leading services. So they have more flexibility with pricing.
Besides, hasn't SCotUS ruled that raw LLM output isn't subject to copyright? So these companies would be breaking a ToS at worst.
I wonder how much of DeepSeek and Xiaomi's pricing cuts can be traced back to cheap energy in China.
I've heard on podcasts that AI data centers in the US are powered by natural gas. Apparently there is currently a glut of natural gas. So the energy costs are actually pretty low in the US.
Earlier quoted context omitted.
So you think they’re running the same types of state of the art Nvidia deployments?
It's supposed to be even MORE expensive: Nvidia H100: Typically priced around $25,000–$30,000 (global MSRP). Huawei Ascend 910C: Reported to cost roughly $28,000, yet it delivers only 60% of the inference performance of the Nvidia H100. Google's TPUs are significantly cheaper for Google for inference. That's pretty much it. There's a reason nVidia has an 80% margin right now.
Earlier quoted context omitted.
There's evidence various third-party models (including Deepseek) used distilling in training, based on models from those leading services. So they have more flexibility with pricing.
So? And Anthropic/OpenAI literally stole copyrighted content to train their models.