DeepSeek V3 0324
planned
M
Michał
This model seems to outperform Claude 3.7 at certain tasks and is the first non-reasoning model to use some mini reasonings within the main answer, which makes it particularly smart (for a non-CoT type of model). It is available on OpenRouter, provided by chutes.ai which seems to host their stuff on US-located servers.
endu
Merged in a post:
DeepSeek V3 0324 upgrade
t
theunmindful
The newly released model is competing for best non-reasoning model out there as benchmarks and user feedback. It can be upgraded here at Merlin
endu
Merged in a post:
🫰 Deepseek v3.1
Daniel Eduardo Martinez Ramirez
Include deepseek 3.1 �
endu
planned
endu
Hi Michal & Monde Mamon! if it's available in APIs, we'll work on integrating it. Appreciate your suggestion! 🚀
monde mamon
Deepseek V3 0324 is also much cheaper, when implementing, is it possible to remove the length constraints that is usually on other AI models?
M
Michał
monde mamon That would be fantastic. In fact, all those models that cost 1 to 5 credits shouldn't be artificially capped and should have their native context window length. I see no reason for Haiku 3.5, DeepSeek V3 (or now V3.1), DeepSeek R1 Slow, GPT-4o Mini, or Gemini 2.0 Flash to be limited, since they cost nothing.