im sure folks have seen it - DeepSeek v3.1 - seems like an ideal model for SambaNova - its BIG, and seems can do a bunch of complex logic that needs the speed of SambaNova ….and yeah gpt-oss is needed but just wondering how hard to update SambaNova DeepSeek V3 to V3.1 ?
2 Likes
It was very interesting way of rolling it out. They only released the base version, not the instruct version as of yet.
Once that releases and we review it, we will definitely look into updating. But based on the base model, they have not changed much in the architecture so it should be quick for us to update on the cloud.
1 Like
Looks like it got released shortly after your comment @vasanth.mohan (instruct version) 
1 Like
And we will be releasing it very soon! https://x.com/SambaNovaAI/status/1959041425963983301
1 Like
Good day.
DeepSeek-V3.1 is a hybrid model that supports both thinking and non-thinking modes. Will SambaNova discontinue hosting R1 0528 since it already covers both reasoning and non-reasoning models?
1 Like
Great question! We are planning to keep all models in service when we launch DeepSeek-V3.1. If we do decide in the future to make it an on request model in the future, we will definitely provide notice as per our depreciation policy.
Are you planning to continue using R1? Would love to understand how you are thinking about R1 vs the 3.1 thinking
Thank you for the confirmation!
Deepseek V3.1 thinking mode appears to be an improvement over R1 0528 in reasoning efficiency. I’m looking forward to trying it out once it becomes available on SambaNova.
1 Like
Live now: DeepSeek-V3.1 Is Live on SambaCloud!
1 Like