@nebiusai Imagine how fast the models would be with prompt caching, an essential feature Nebius still don't have, which makes it nearly unusable / to expensive for agentic cases ๐ฅฒ
@xmgplus DeepSeek PRO GA will be insane. The question is: Will GLM 5.3 have vision on board and stay at a similar price.
Kimi K3 is just to expensive and to slow, it could be a great model, but at the end its more expensive then GPT Sol on a lower effort because it needs more turns.
The best models change. Your privacy doesnโt.
The worldโs leading AI models. Always up to date.
Zero Data Retention. No training on your data.
All in Lurus.
Insane model! Tested it and it is saving a lot of money. Can't wait for the pro version.
It reminds me of waiting for Fable, you don't want anything else anymore. If Flash is already this good, where does Pro fit in? Is it the license that will lower its token costs to 10% in the future? Thanks, DeepSeek! The new DeepSeek Flash is now available on Lurus Code.
@thsottiaux Really nice!
Luna probably offers the best value for money of any model worldwide right now. Hopefully Microsoft Azure and Bedrock will follow suit quickly (It will probably take Azure 3 weeks to make this change ๐ฅธ).
I have another suggestion: -10% on Sol? ๐
@sama Really don't know how this is working in 365 Copilot. Azure Foundry GPT 5.6 Series does not return any prompt cache at the moment. Its broken. We are waiting since yesterday for a fix. Unable to use it at the moment for agentic coding ๐ซค.
@elonmusk Congratulations Elon ๐. Europe is waiting for a GDPR-compliant version of Grok 4.5 for company's with Zero Data Retention option available