DeepSeek V4.1 Flash 1017 runs at 1,521+ tok/s on Fireworks Ultrafast. Fast enough to ship Tetris in 11 seconds, 7x faster than Claude Opus 5.5. Join the Ultrafast waitlist: bit.ly/3UA0wKq

Oct 21, 2026 路 6:27 PM UTC

80
157
85
2,164
256,952
RelevantRecentLikes
Crazy stuff. Even crazier to think we鈥檙e still in the 56 kbps era of AI
3
3
104
7,073
as long as you dont surreptitiously farm my data on train on my data, let's use this
1
1,285
maybe im not working hard enough cause i cant think of a usecase in my life where i need a model to run this fast
20
2
1
64
13,080
I'm very interested to see the price for this. And find out what hardware it runs on 馃憖
12
On how much vram?
242
Well. That is fast.
2
4,960
bro get me out of waitlist please
503
I need 10kt/s 馃お馃槀
33
What about cost per task?
3
27
11,841
Is the infra behind this still @OracleCloud
1
3,074
is the wait list ultrafast?
7
5,282
Will this be available on openrouter?
1
5
5,722
@GilFoss imagine how many use cases this opens up
2
3,292
@bot notify me when there鈥檚 an intelligence benchmark on this already
4
4,433
Can you get the long piece to show up in 11 seconds too?
1,135
pricing?
371
Tetris in 11 seconds is a fun speed demo. I'd judge the coding claim by whether the generated code passes a test suite.
2,560
EdDiboi
@eddiboi13h
what's the pricing?
1
2,448
@bot sign me up like yesterday
1
1,892
This is badass... So many ideas 馃馃檭
1,704
why 4.1 flash reasong off is there a particular reason for that?
2,206
馃く
2,749
So much ideas i have depend on the time when we have LLMs with such TPS! Look forward to testing it!
2
GOING TOO FAST
1
621
High throughput is useful, but practical comparisons also need the workload, output quality and serving cost alongside tokens per second. Those details help teams judge whether speed changes the user experience.
1
151
tetris in 11 seconds is fun but the real win at 1,500 tok/s is iteration. you can afford to throw away 10 attempts and keep the best one in the time a slow model writes one
10
2,722
crazy
4
what a comparison reasoning off on deepseek vs default reasoning on opus 5.5 ( Very fair comparison) 馃憤馃徎
68
Crazy
115
Crazyyy
40
Price?
386
Holyyyy
172
@FireworksAI_HQ
652
wooooow
1
612
what !!
199
Signed up!
150
Load more replies