@gigaqian: DeepSeek just released a new version of their V4 Flash model with a massive upgrade in agentic performance! They claim that all these improvements came from “re-post-training,” meaning that nothing about the size or architecture of the model changed at all. This could signal another massive moment for DeepSeek as they prepare to release the newest version of their flagship model, DeepSeek V4 Pro! #deepseek #ai #aitraining #ainews #gigaqian
I love DeepSeek.it is so cool and I use it to learn simple codes.
2026-08-19 21:43:25
0
Tony :
Use them (DeepSeek v4 Flash/Pro) a lot between limits of Claude and Codex, both super capable at low cost.
2026-08-03 05:03:18
5
Nina Mar :
They call this the deepseek kill zone because of how intelligent and cheap it is.
2026-08-03 01:38:35
41
Tylers Top Trends :
As a normal.. would be cool to understand how they actually “train” these things and make them better
2026-08-03 02:25:20
6
ntncy :
Thanks for the constant updates flow, keep them coming 👌👌👌
2026-08-02 23:33:14
17
Johnohhh1 :
i think so many labs are all figuring out something similar. These teams are really winning
2026-08-03 00:29:38
9
FeaturingBatty :
Do you recommend any Chinese model as a viable replacement for Claude Code? K3 is slow and I’m wondering if there’s someone I’m missing? 👀 If not, still love the content!
2026-08-03 06:11:30
3
RJ Gracias :
very wise to include date on your content🤌
2026-08-02 23:35:01
5
Mark Cooper :
I feel there needs to be new benchmarks to compare models and performance of the frontier models and capabilities.
2026-08-03 01:07:59
1
瑶台镜月色寒 :
2026-08-03 08:01:43
10
Young money :
was kimi3 weight released to hugging face?
2026-08-03 20:35:27
2
Brian 🇺🇸 :
DeepSeek v4 Pro is already a very capable (and cheap) coder.
2026-08-03 03:44:21
1
John Peabody :
Deep seek risk score so high it has to be good!
2026-08-02 23:58:11
1
Jesse Campbell :
it doesn't seem to be bench maxed, sounds like it's the real deal
2026-08-03 11:41:09
1
BlackN⚫va :
thank you for posting the date!!!!!
2026-08-03 04:19:37
1
THE ONE AND ONLY :
Awesome info
2026-08-02 23:53:43
1
source🇨🇦🇺🇲 :
Same model same size but radicaly radicaly different processing
You are are welcome.
There is alot going on behind the scenes in ai that most people will never even touch.
2026-08-04 00:57:42
1
god :
now you know openai and claude are all scam
2026-08-03 14:16:46
1
Jabadiya :
it's called fine tuning
2026-08-03 08:00:19
2
Sir Zeno :
2026-08-04 12:17:56
0
Pablo :
Do I need a new API key or can I just keep using the previous one?
2026-08-04 18:13:28
0
FiveClaw :
Deepseek flash is insanely good like Opus level
2026-08-05 20:56:35
0
666bbb.0 :
Deep Sneak. Get the name correct🤣
2026-08-05 05:49:36
0
Tiktrending :
thanks for the info
2026-08-11 17:47:59
0
George Patience242 :
wait till you get a load of DeepSeek Cephalopod 🐙🤯🤫
2026-08-04 04:12:49
0
To see more videos from user @gigaqian, please go to the Tikwm
homepage.