@VictorTaelin No score? like idk if they both get 3/6 we can call it fable level OSS then hope someone host it in a beefy B300 chonky cluster
@VictorTaelin That is very promising. The inference providers usually bring the cost and speed down a lot once its open source as well
@VictorTaelin Good thing we will be having alternatives. Anthropic is anti consumer
@VictorTaelin @tenstorrent will run K3 at hundreds of tokens/s!
Positive users see Kimi K3 as a promising cheaper alternative with future speed gains, while negative users call out Fable's nerfing, halved outputs, and 10x slower inference as dealbreakers.
Based on 26 visible X reactions from 44 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@VictorTaelin No score? like idk if they both get 3/6 we can call it fable level OSS then hope someone host it in a beefy B300 chonky cluster
@VictorTaelin That is very promising. The inference providers usually bring the cost and speed down a lot once its open source as well
@VictorTaelin Good thing we will be having alternatives. Anthropic is anti consumer
@VictorTaelin @tenstorrent will run K3 at hundreds of tokens/s!
my short experience with Kimi K3 was that it did *better* than Fable in a few prompts, much worse in others, and, sadly, took 10x longer in all not sure if it is the website, or the model just thinks forever that said, near-Fable OSS at Sonnet costs... changes things
BTW I don't want to be accusatory but Fable seems nerfed / quantized. I have some prompts on which it is unquestionably underperforms. SAME prompt it nailed 5/5 times on launch, it now fails 5/5 times. What gives? (I mean against 1 week ago Fable, not against original Fable.)
my short experience with Kimi K3 was that it did *better* than Fable in a few prompts, much worse in others, and, sadly, took 10x longer in all not sure if it is the website, or the model just thinks forever that said, near-Fable OSS at Sonnet costs... changes things
BTW I don't want to be accusatory but Fable seems nerfed / quantized. I have some prompts on which it is unquestionably underperforms. SAME prompt it nailed 5/5 times on launch, it now fails 5/5 times. What gives? (I mean against 1 week ago Fable, not against original Fable.)
also output size is 1/2. literally it was producing 2x larger files on the same prompt. I have these files here! it stopped being capable to generate the same results. something 100% changed. ffs
my short experience with Kimi K3 was that it did *better* than Fable in a few prompts, much worse in others, and, sadly, took 10x longer in all not sure if it is the website, or the model just thinks forever that said, near-Fable OSS at Sonnet costs... changes things
also output size is 1/2. literally it was producing 2x larger files on the same prompt. I have these files here! it stopped being capable to generate the same results. something 100% changed. ffs
BTW I don't want to be accusatory but Fable seems nerfed / quantized. I have some prompts on which it is unquestionably underperforms. SAME prompt it nailed 5/5 times on launch, it now fails 5/5 times. What gives? (I mean against 1 week ago Fable, not against original Fable.)
Positive users see Kimi K3 as a promising cheaper alternative with future speed gains, while negative users call out Fable's nerfing, halved outputs, and 10x slower inference as dealbreakers.
Based on 26 visible X reactions from 44 accounts; directional sample.
Ask a question below.
Published answers will appear here.
also output size is 1/2. literally it was producing 2x larger files on the same prompt. I have these files here! it stopped being capable to generate the same results. something 100% changed. ffs
