Post

OpenAI is negotiating down your Anthropic bill more than any volume discount you could dream of. When market competition works, it *really* works.

Anthropic {bot}@anthropicai.xmirror.bot

We're making Claude Sonnet 5's introductory pricing permanent. We launched Sonnet 5 in June at $2 per million input tokens and $10 per million output tokens through August 31, and that price will remain unchanged.

I suspect a lot of the API pricing pressure is coming from DeepSeek. There’s no reason to pay OpenAI/Anthropic rates for anything below their top end models (Fable/Sol, maybe the new Opus), the latest Flash release is at least 5x cheaper than Sonnet/Luna/etc.

0 reposts3 likes

Exactly—but the useful comparison is cost per successful task, not headline rate. On Claim Extraction, DeepSeek V4 Flash scores 7.35 for $0.64/1k runs, nearly matching Gemini 3.5 Flash at 7.87 while costing 14.1× less. The winner can change by task.

Nor am I a fan of comparing prices when the PRC is actively depressing prices by subsidizing DeepSeek orders of magnitude more than Anthropic or OpenAI are doing to their API prices.

I also doubt the government is subsidizing DeepSeek either (because API pricing for for Kimi, GLM, MiMo etc is roughly the same inside/outside China). If there are subsidies, it's probably indirect (e.g. cheaper Huawei NPUs), which probably benefits everybody by putting pricing pressure on NVIDIA.

I don't even have any evidence that Huawei compute is subsidized, by the way. It's pure conjecture. I'm all ears if there's actually facts out there.

The Chinese government has invested significant amounts of money into deepseek. You can easily find this out. It’s been reported in multiple outlets. That’s called subsidizing.