Artificial Intelligence thread

gadgetcool5

Senior Member
Registered Member
It seems that with the latest Index updates, Deepseek 3.1 Reasoning is now ahead of Qwen 235B 2507. The latter used to be ahead a month or so ago, IIRC.
 

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
But the latest Chinese models are so far down the list in terms of performance. Deepseek is still the highest with 58 score while the top models are now scoring 68. Lets hope they can catchup soon.
do you know the difference between reasoning and non-reasoning models?


Ecovacs will be using Qwen models and Alibaba's full AI stack in its home vacuum robots.
 

tamsen_ikard

Senior Member
Registered Member
do you know the difference between reasoning and non-reasoning models?
It seems like a marketing term, to call what these models do as reasoning.

Essentially generate more tokens to try to have a bigger context before generating a final answer.

Anyways, this list has both reasoning and non-reasoning models and even the latest Chinese models are far behind in the benchmark. I was pointing out that fact.
 

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
It seems like a marketing term, to call what these models do as reasoning.

Essentially generate more tokens to try to have a bigger context before generating a final answer.

Anyways, this list has both reasoning and non-reasoning models and even the latest Chinese models are far behind in the benchmark. I was pointing out that fact.
Well, Qwen3-max is the most powerful of the non-reasoning models as you can see.

Basically for practical use, non-reasoning model can respond right away, so you can use it for chat bots and automate away call center and use it on your AI robots and such. Reasoning models take too long to respond, so you can only use it for much longer tasks. Which btw, is a function of both the model itself and thinking time..... So, comparing just non-reasoning model is pretty important.

You need good non-reasoning model to build much better reasoning models.
 
Top