Is this another DeepSeek moment? Open-source passing closed-source again Should we expect this every couple months now?
Kimi.ai
 @Kimi_Moonshot
Nov 6
🚀 Hello, Kimi K2 Thinking! The Open-Source Thinking Agent Model is here. 🔹 SOTA on HLE (44.9%) and BrowseComp (60.2%) 🔹 Executes up to 200 – 300 sequential tool calls without human interference 🔹 Excels in reasoning, agentic search, and coding 🔹 256K context window Built as a thinking agent, K2 Thinking marks our latest efforts in test-time scaling — scaling both thinking tokens and tool-calling turns. K2 Thinking is now live on kimi.com in chat mode, with full agentic mode coming soon. It is also accessible via API. 🔌 API is live: platform.moonshot.ai 🔗 Tech blog: moonshotai.github.io/Kimi-K2… 🔗 Weights & code: huggingface.co/moonshotai

Nov 6, 2025 · 5:03 PM UTC

75
78
7
1,140
Replying to @Thom_Wolf
There’s no point implementing these models at scale until development properly plateaus. Like, why bother with K2 when Gemini 3 drops soon and you’ll want to experiment with that instead
1
Made in the USA NDAA Compliant AI NVIDIA Flight Controllers. Ready to scale your autonomous drone production.

ARK Jetson Flight Controller

arkelectron.com
3
8
3
96
Replying to @Thom_Wolf
DeepSeek never surpassed frontier models other than some meaningless benchmarks, but as someone who uses regular K2 frequently, I believe its thinking model might actually surpass them.
2
7
Replying to @Thom_Wolf
This is the much needed competition, else Anthropology and ClosedAI will bleed us dry. Chinese Open Source AI will humble these American hype-masters.
1
36
This tweet is unavailable
Replying to @Thom_Wolf
For agentic stuff we have to wait to see if the reality matches these benchmarks. that being said, the model has been really impressive so far
2
6
Replying to @Thom_Wolf
It’s starting to feel like open-source projects wake up every few months and say, “Alright, let’s humble the giants again.” The pace is wild, almost like a tech version of tag where the underdog keeps catching the mega corporations off guard. At this rate, even the closed systems might turn open just to survive.
1
Replying to @Thom_Wolf
It can’t go like this all the time, they’ll become so good at some point that the question will be “what’s next?”
Replying to @Thom_Wolf
No I don't think it is. I'm playing with it today and it's not all that performant... not in practice.
1
Replying to @Thom_Wolf
Writing is on the wall for those who are paying attention
Replying to @Thom_Wolf
Nation state subsidized.
3
1
2
Replying to @Thom_Wolf
Kimi VLA pls 🙏
1
Replying to @Thom_Wolf
Yes! Let’s go! 🚀
1
Replying to @Thom_Wolf
Deepseek wasn't a moment, it was a small ripple and merely an echo of real work.
1
1
Replying to @Thom_Wolf
open-source leapfrogging closed models is happening way faster than anyone thought. performance gap closing every cycle, kimi ai stats speak volumes, but the hardest problems still need a reality check.
Replying to @Thom_Wolf
If @perplexity_ai swaps out R1 to K2 in sonar that would indicate that is better and not so much more expensive
Replying to @Thom_Wolf
I think there will come a point where the open source is so close and the performance is more than enough that we will be at a point where closed source is hanging on by the brand name, or perhaps UX. Currently, the closed-source companies are racing towards model performance and forgoing features and usability.
4
Replying to @Thom_Wolf
There is so much happening in China, even beyond classic open source models. Applied AI (robotics, manufacturing, etc) will be very exciting and where open sourcing could really have a huge impact.
1
Replying to @Thom_Wolf
it's amazing how chinese labs do this with so little..
This tweet is unavailable
Replying to @Thom_Wolf
Does it rat you out to Xi tho
1
Replying to @Thom_Wolf
We built a hacker agent and used it to benchmark all models on Openrouter, Kimi K2 Thinking scores really well across Factual accuracy, Prompt resistance and Data protection. So strong guardrails built into the model
Replying to @Thom_Wolf
love it cause the big players gotta do big players stuff now
Replying to @Thom_Wolf
Looks like china wants the AI to be decentralized by open sourcing everything.
Replying to @Thom_Wolf
Interesting! Have you tried their model yet?
Replying to @Thom_Wolf
Collaboration wins.
Replying to @Thom_Wolf
That's what the singularity is.
Replying to @Thom_Wolf
Probably even more frequently until it just will be always better
5
Replying to @Thom_Wolf
Between Kimi, MiniMax and Qwen.. and hopefully DeepSeek... open source could remain steady competition
Replying to @Thom_Wolf
So interesting watching this all play out :) love the resurgence of open source with AI
Replying to @Thom_Wolf
As long as open-source is at most 1cyear behind closed source, I'm happy
Replying to @Thom_Wolf
feels like someone uploaded the weights of agi
Replying to @Thom_Wolf
There will be no more DeepSeek moments.
1
Replying to @Thom_Wolf
I think we’ll start to see this leapfrogging more often. Open source is getting really good - Realistically, it’s still anyone’s game.
Replying to @Thom_Wolf
The Kimi moment
Replying to @Thom_Wolf
let Open-source great again
Replying to @Thom_Wolf
I love open-source but,
8
Load more