@olleai: Opus 4.8 and GPT-5.5 both run nearly 40x slower than this model on Cerebras. This kind of throughput is only possible via smaller open models because of their size and availability to other providers. #gptoss #opensourceai

Olle
Olle
Open In TikTok:
Region: CH
Saturday 20 June 2026 18:56:14 GMT
13165
438
19
51

Music

Download

Comments

cannotgoback
亚当(Adam回不去了,欧米加版) :
oss120 verybad can’t event use tool call properly
2026-06-21 05:41:22
6
philippe.bourque
Philippe Bourque :
500 Tok/sec on 5090 for qwen3.6-35b ✌️
2026-06-22 01:08:37
5
pirkkapekkapeteli
pirkkapekkapeteli :
How is the accuracy with those speeds? How did you generate the plan for that project
2026-07-10 19:03:13
1
darksideavatar
DarkSideAvatar :
Just filtered by provider , it only has gpt-oss-120b and glm4.7 , unless it gets newer models speed won’t mean much
2026-06-21 02:56:36
1
luckyranasinghe56
luckyranasinghe56 :
What provider you use
2026-07-29 05:42:52
0
dekitvasyl
Kit Vasyl :
generator faster than compile
2026-06-20 19:33:34
3
shrodkx
. :
can it work on glm and kimi models
2026-08-06 01:22:12
0
pinkola52
pinkola :
bro big companies they give you 60 tokens per second so they don't flood their system they can you type and results in front of you
2026-06-25 15:35:21
1
_emmy_alt
✨EMMA✨❤️‍🔥💓💓(simpatic) :
proofademic flagged text i wrote using the recommended thesis structure
2026-06-22 03:52:48
0
sveinbjornp
Sveinbjörn :
Is it useful for anything though?
2026-06-22 13:41:20
0
gal.levinsky
Gal Levinsky :
thanks for the info!
2026-06-20 19:24:12
0
jimjimmabi
JimjimMabi :
betalar han eller köra han ai på hans dator?
2026-06-25 12:35:04
0
krieg84
Krieg :
Gpt 5.6 sol Is running at 650 tokens per second
2026-06-29 16:46:55
0
To see more videos from user @olleai, please go to the Tikwm homepage.

Other Videos


About