#k2think - Twitter Hashtag

5 months ago

We have submitted our application for K2 Think V2! We want to integrate K2 as a reasoning module into our on-device cognitive architecture based on Active Inference - fully local, sovereign, for HSE and smart cities. Open for collaboration! #K2Think #MBZUAI #SovereignAI

ArchotecAI's tweet photo. We have submitted our application for K2 Think V2! We want to integrate K2 as a reasoning module into our on-device cognitive architecture based on Active Inference - fully local, sovereign, for HSE and smart cities. Open for collaboration! #K2Think #MBZUAI #SovereignAI https://t.co/2F3FlBA8Ch

MBZUAI

@mbzuai

5 months ago

If your project can think, we want to see it 👀 Build with K2 Think V2, submit your demo, and get a shot at rewards up to $30K and real visibility. New round every month. First deadline: March 10, 2026 https://t.co/WexIHKtIgI

mbzuai's tweet photo. If your project can think, we want to see it 👀

Build with K2 Think V2, submit your demo, and get a shot at rewards up to $30K and real visibility.

New round every month.

First deadline: March 10, 2026

https://t.co/WexIHKtIgI https://t.co/m08qBPxnHT

2

12

1

5

810

0

2

0

55

ArchotecAI @ArchotecAI

5 months ago

Agree with @mbzuai on AI sovereignty - key for trust! Archotec AI builds on this with local on-device Active Inference for robust real-world apps in UAE HSE/smart cities. What do you think? #SovereignAI #ResponsibleAI #K2Think @MBZUAI @Hub71

MBZUAI

@mbzuai

5 months ago

Sovereignty matters when AI systems start to matter. As AI systems move from labs into institutions, the question shifts from “how well does it perform?” to “what are we actually deploying?” With K2 Think V2, sovereignty means owning the full lifecycle of the model without hidden dependencies or external assumptions. This matters because long-term trust in AI systems relies on independence, accountability, and the ability to stand behind the system you deploy. K2 Think V2 reflects our belief that sovereign AI foundations are not a future concern, but a present requirement. Download here: https://t.co/4ccNrTsDOf https://t.co/8ODkcST9ne #AISovereignty #ResponsibleAI #FoundationModels #K2Think #MBZUAI

1

10

1

0

819

0

3

0

24

Eric Xing

@ericxing

5 months ago

The #K2Think V2 we released today represents a significant leap of the already-strong performance on deep reasoning over V1 (https://t.co/qiELrJOmDt), and furthermore drastically improves on reliability, reducing the hallucination rate from 89% to 52% on the AA-Omniscience benchmark while simultaneously increasing long-context reasoning accuracy from 33% to 53%. Powered by the K2-V2 Instruct foundation, it is now the strongest fully open-source reasoning model available, and rivals all frontier reasoning models from the same size class, open or close. Continuing the founding ethos of Institute of Foundation Models (IFM) at @mbzuai, we open source the weights, data, recipe, and development of this release: on top of the #K2Think V1 recipe, and the strong foundation of K2 V2 (https://t.co/0Kymo2B0mI) base LLM, we further made a few tailored designs that adapt well with the upgraded base model: 1) GRPO with asymmetric policy ratio clipping 2) expanded high quality STEM data sources 3) capability steering with model specific difficulty-based data filtering, 4) full on-policy training with large batch sizes. 5) two-stage progressive context length expansion for training efficiency. Check out our blog post and model for more details: https://t.co/gRQCevpzi0 @waterluffy, @llm360_agi, @allen_ai, @huggingface

ericxing's tweet photo. The #K2Think V2 we released today represents a significant leap of the already-strong performance on deep reasoning over V1 (https://t.co/qiELrJOmDt), and furthermore drastically improves on reliability, reducing the hallucination rate from 89% to 52% on the AA-Omniscience benchmark while simultaneously increasing long-context reasoning accuracy from 33% to 53%. Powered by the K2-V2 Instruct foundation, it is now the strongest fully open-source reasoning model available, and rivals all frontier reasoning models from the same size class, open or close.

Continuing the founding ethos of Institute of Foundation Models (IFM) at @mbzuai, we open source the weights, data, recipe, and development of this release: on top of the #K2Think V1 recipe, and the strong foundation of K2 V2 (https://t.co/0Kymo2B0mI) base LLM, we further made a few tailored designs that adapt well with the upgraded base model: 1) GRPO with asymmetric policy ratio clipping 2) expanded high quality STEM data sources 3) capability steering with model specific difficulty-based data filtering, 4) full on-policy training with large batch sizes. 5) two-stage progressive context length expansion for training efficiency. Check out our blog post and model for more details: https://t.co/gRQCevpzi0

@waterluffy, @llm360_agi, @allen_ai, @huggingface

Artificial Analysis

@ArtificialAnlys

5 months ago

MBZUAI’s Institute for Foundation Models’ new K2 Think V2 model improves on 70B K2-V2 in intelligence, maintains its tied first place in our Openness Index, and has a low hallucination rate 🇦🇪 Third UAE model on Artificial Analysis in the last month: K2 Think V2 is a dense 70B reasoning model. It was post-trained from MBZUAI’s K2-V2, previously released in December 2025. We also recently evaluated Falcon-H1R-7B from TII, another government-sponsored research center from the UAE 📖 Pushes the frontier in Intelligence vs Openness: Like its base model, K2-V2, K2 Think V2 maintains its tied lead in Openness with the Olmo family from Allen Institute for AI. The Artificial Analysis Openness Index is our newly released, standardized, independently assessed measure of AI model openness across availability and transparency. K2 Think V2 provides full access to pre- and post-training data, as well as publishing training methodology and code with a permissive Apache license allowing free use for any purpose. The model is the most intelligent model among highly open peers 🔮 Low hallucination rate: K2 Think V2’s hallucination rate is among the lowest of relevant comparison models. Its hallucination rate sits between Anthropic’s Claude 4.5 Opus and Claude 4.5 Sonnet - Anthropic being a leader in low-hallucination models. The Artificial Analysis Omniscience Hallucination Rate highlights how often models will incorrectly answer knowledge-based questions, rather than abstain. The significant improvement from K2-V2’s hallucination rate from 89% to 52% demonstrates how post-training can effectively reduce hallucination. This contributes to a strong score of -26 for K2 Think V2 in AA-Omniscience, our new knowledge and hallucination benchmark 📈 Improved intelligence among medium-sized (40-150B) open weights models: K2 Think V2 improves by 4 points on its base model. This improvement is driven largely by its reduced hallucination rate (described above), but also its improved long context reasoning capability. K2 Think V2 scores 53% in AA-LCR, our long context reasoning benchmark, a significant jump from 33% in K2-V2 ⚡Slightly improved token efficiency: K2 Think V2 uses 10% fewer tokens (99M) than its base model (110M) to run our Intelligence Index Congratulations to the team at @mbzuai !

ArtificialAnlys's tweet photo. MBZUAI’s Institute for Foundation Models’ new K2 Think V2 model improves on 70B K2-V2 in intelligence, maintains its tied first place in our Openness Index, and has a low hallucination rate

🇦🇪 Third UAE model on Artificial Analysis in the last month: K2 Think V2 is a dense 70B reasoning model. It was post-trained from MBZUAI’s K2-V2, previously released in December 2025. We also recently evaluated Falcon-H1R-7B from TII, another government-sponsored research center from the UAE

📖 Pushes the frontier in Intelligence vs Openness: Like its base model, K2-V2, K2 Think V2 maintains its tied lead in Openness with the Olmo family from Allen Institute for AI. The Artificial Analysis Openness Index is our newly released, standardized, independently assessed measure of AI model openness across availability and transparency. K2 Think V2 provides full access to pre- and post-training data, as well as publishing training methodology and code with a permissive Apache license allowing free use for any purpose. The model is the most intelligent model among highly open peers

🔮 Low hallucination rate: K2 Think V2’s hallucination rate is among the lowest of relevant comparison models. Its hallucination rate sits between Anthropic’s Claude 4.5 Opus and Claude 4.5 Sonnet - Anthropic being a leader in low-hallucination models. The Artificial Analysis Omniscience Hallucination Rate highlights how often models will incorrectly answer knowledge-based questions, rather than abstain. The significant improvement from K2-V2’s hallucination rate from 89% to 52% demonstrates how post-training can effectively reduce hallucination. This contributes to a strong score of -26 for K2 Think V2 in AA-Omniscience, our new knowledge and hallucination benchmark

📈 Improved intelligence among medium-sized (40-150B) open weights models: K2 Think V2 improves by 4 points on its base model. This improvement is driven largely by its reduced hallucination rate (described above), but also its improved long context reasoning capability. K2 Think V2 scores 53% in AA-LCR, our long context reasoning benchmark, a significant jump from 33% in K2-V2

⚡Slightly improved token efficiency: K2 Think V2 uses 10% fewer tokens (99M) than its base model (110M) to run our Intelligence Index

Congratulations to the team at @mbzuai !

7

99

25

23

19K

2

22

4

3K

قناة المال | Almal Channel

@Almalchannel

8 months ago

تنبيه: في لاعب جديد دخل ملعب الذكاء الاصطناعي اسمه K2 Think ومن صنع الإمارات 🇦🇪 الهدوء اللي قبل العاصفة... أحياناً يكون في أبوظبي #K2Think #ذكاء_اصطناعي #الإمارات #تقنية

11

86

46

20

12K

Digital Inside España @digitalinsidees

10 months ago

Según sus promotores, el nuevo modelo K2 Think supera a modelos mucho mayores gracias a una combinación de nuevas técnicas de entrenamiento e inferencia, y ya se encuentra disponible a través de un chatbot online. 👉 https://t.co/FcCWZuJRMj #K2THINK #MBZUAI

digitalinsidees's tweet photo. Según sus promotores, el nuevo modelo K2 Think supera a modelos mucho mayores gracias a una combinación de nuevas técnicas de entrenamiento e inferencia, y ya se encuentra disponible a través de un chatbot online.

👉 https://t.co/FcCWZuJRMj

#K2THINK #MBZUAI https://t.co/twVTe5E4Cj

0

2

1

0

90

سيف محمد بن صفوان | Saif Bin Safwan @Saif_BinSafwan

10 months ago · United Arab Emirates

إطلاق نموذج #K2Think يمثل نقلة نوعية تعكس طموح #الامارات لترسيخ مكانتها مركزاً عالمياً للذكاء الاصطناعي بدعم من أحدث التقنيات وتدريب محلي رفيع إنها ليست خطوة تقنية فقط، بل إعلان عن ريادة ورؤية استراتيجية تقود ثورة #الذكاء_الاصطناعي عالمياً. https://t.co/WgqEEQqHwK

0

2

0

304

UAE Forsan

@UAE_Forsan

10 months ago

K2 Think: A Breakthrough in Advanced AI Reasoning, Reinforcing UAE’s Global AI Leadership #UAE #AbuDhabi #AI #K2Think #MBZUAI #G42 @mbzuai @G42ai @ADMediaOffice

UAE_Forsan's tweet photo. K2 Think: A Breakthrough in Advanced AI Reasoning, Reinforcing UAE’s Global AI Leadership
#UAE #AbuDhabi #AI #K2Think #MBZUAI #G42
@mbzuai
@G42ai
@ADMediaOffice https://t.co/UgZTzSODTk

0

3

2

0

244

Erfan Zolfaghari @erfanz_eth

10 months ago

🤯 مدل K2-Think از امارات با فقط ۳۲ میلیارد پارامتر، عملکرد GPT-4 رو می‌زنه و ۲۰ برابر کوچکتره! سرعت ۲۰۰۰ توکن/ثانیه روی سخت‌افزار عمومی. دیگه نیازی به مدل‌های غول‌پیکر نیست؟ #AI #K2Think #هوش_مصنوعی

1

4

0

1

341

محتوى @Mohtawaae

10 months ago

#الإمارات تطلق #K2Think.. النموذج المفتوح الأكثر تقدماً عالمياً في الاستدلال بـ#الذكاء_الاصطناعي.. بتعاون بين #MBZUAI وشركة #G42.. ليقدم أداءً يضاهي نماذج تفوقه حجماً بـ10 مرات 🌍🤖 #شغف_بالمعرفة

0

1

0

811

فرسان 71

@Forsan_71

10 months ago

تخيّل أداة تفكّر وتستدل مثل العقل البشري بس بسرعة البرق! ⚡️ هني يي دور K2 Think… مشروع إماراتي ذكي يفتح باب جديد للفكر والتحليل. #K2Think #الإمارات #ابتكار

0

5

24

0

3K

Manu Kumar Jain

@manukumarjain

10 months ago

#K2Think - scaling another “peak” of innovation! 🚀 In the #AI world, scale has long been seen as the path to intelligence. But with K2 Think, we’re showing the world that smarter can be better than bigger. Developed by #MBZUAI and #G42, this 32B open-source model is compact yet remarkably powerful and at par with much larger peers in complex reasoning. Its design goes beyond prediction to true intelligence: * Long chain-of-thought supervision * Reinforcement learning with verifiable rewards * ⁠Agentic planning & speculative decoding * ⁠Test-time scaling for precision reasoning Powered by Cerebras Systems’ breakthrough compute, K2 Think runs at lightning speed, up to 2,000 tokens per second. ⚡ ⚡ K2 Think has already established itself among the industry’s top reasoning systems, delivering scores that tie it for top math performance across benchmarks. K2 Think is more than a model. It is the future. This is proof that future of reasoning will be shaped not only by size, but by ingenuity and collaboration. Learn more: https://t.co/RqysMRTLvO @G42ai #ArtificialIntelligence #AI #K2 #Think #UAE #AbuDhabi

manukumarjain's tweet photo. #K2Think - scaling another “peak” of innovation! 🚀

In the #AI world, scale has long been seen as the path to intelligence. But with K2 Think, we’re showing the world that smarter can be better than bigger.

Developed by #MBZUAI and #G42, this 32B open-source model is compact yet remarkably powerful and at par with much larger peers in complex reasoning.

Its design goes beyond prediction to true intelligence:
* Long chain-of-thought supervision
* Reinforcement learning with verifiable rewards
* ⁠Agentic planning & speculative decoding
* ⁠Test-time scaling for precision reasoning

Powered by Cerebras Systems’ breakthrough compute, K2 Think runs at lightning speed, up to 2,000 tokens per second. ⚡ ⚡

K2 Think has already established itself among the industry’s top reasoning systems, delivering scores that tie it for top math performance across benchmarks.

K2 Think is more than a model. It is the future. This is proof that future of reasoning will be shaped not only by size, but by ingenuity and collaboration.

Learn more: https://t.co/RqysMRTLvO

@G42ai #ArtificialIntelligence #AI #K2 #Think #UAE #AbuDhabi

2

30

4

2

4K

MONIIFY

@MoniifyBusiness

10 months ago

Abu Dhabi just launched an AI model smarter than its size. K2 Think = lean, fast, open-source genius 🧠🇦🇪 @mbzuai @G42ai #K2Think #AI #OpenAI #ChatGPT #DeepSeek

2

25

4

7

12K

Age of AI Tools @ageofaitools

10 months ago

🇦🇪 UAE just entered the AI chat with K2 Think! 32B parameters delivering OpenAI-level performance at a fraction of the cost. Open source = game changer. The AI race just got more interesting! 🚀 Who's next to surprise us? #AI #K2Think #OpenAI #Innovation

$ageofaitools's tweet photo. 🇦🇪 UAE just entered the AI chat with K2 Think! 32B parameters delivering OpenAI-level performance at a fraction of the cost. Open source = game changer. The AI race just got more interesting! 🚀 Who's next to surprise us? #AI #K2Think #OpenAI #Innovation https://t.co/8hzwr7RvZQ$

0

2

0

134

Omar Al Attar @o_al36ar

10 months ago

Sheikh Khaled once said the UAE “punches more than our weight.” K2 Think proves it: a 32B open-source reasoning model from MBZUAI & G42 that solves math & finance with clarity models 20x larger can’t match. Reasoning > size. #AI #UAE #K2Think

o_al36ar's tweet photo. Sheikh Khaled once said the UAE “punches more than our weight.”
K2 Think proves it: a 32B open-source reasoning model from MBZUAI & G42 that solves math & finance with clarity models 20x larger can’t match.
Reasoning > size.
#AI #UAE #K2Think https://t.co/B3qL2Ne5Uz

1

2

0

162

ARCHETYP Staffing @archetyp_staff

10 months ago

Staffing Magazine更新！本日は、「毎秒2000トークン！K2 Thinkが証明した小型AIモデルの可能性」について解説しています。ぜひチェックしてみてください！ #K2Think #生成AI #MBZUAI #AIProfessional https://t.co/DruMusH50G

0

2

0

278

Taylor W. Killian @tw_killian

10 months ago

We've been really fortunate to partner with @cerebras in the deployment of #K2Think. Rapid, large scale reasoning made possible with their Wafer-Scale Engine. Give it a spin at https://t.co/3QVlEE9MfQ

Cerebras

@cerebras

10 months ago

K2 Think is now available and runs fastest on Cerebras Inference at 2,000 TPS on Cerebras – 20x faster than GPU. Launched by @mbzuai and @G42ai, K2 Think is a world leading open-source reasoning model that beats GPT OSS 120B and DeepSeek on math and reasoning.

27

312

40

84

31K

1

3

0

1

825

MBZUAI

@mbzuai

10 months ago

No sign-ups, no drama. 🎉 Just pure speed. See K2 Think in action and experience advanced AI reasoning at https://t.co/ltN5Yxmtvr - now with no registration required. #K2Think #AI #OpenSource #MBZUAI #G42 #Innovation

4

43

9

5

5K

قناة المال | Almal Channel

@Almalchannel

10 months ago

تنبيه: في لاعب جديد دخل ملعب الذكاء الاصطناعي اسمه K2 Think ومن صنع الإمارات 🇦🇪 الهدوء اللي قبل العاصفة... أحياناً يكون في أبوظبي #K2Think #ذكاء_اصطناعي #الإمارات #تقنية

0

16

13

3

3K

Anna MacMillan

@annamacmillan

10 months ago

Thoughts? Is the future of AI in efficiency rather than endless scaling? Drop your takes below! #K2Think #OpenSourceAI #xAI

0

5

0

71

Web3Aible

@Web3Aible

10 months ago

I tested #K2Think by @mbzuai and in one shot, it crushed the most challenging Quest2 in just 7 seconds! Note only 03, 04mini, gpt-5 (mini-High, High, High-new-system-prompt), Grok 3 mini-High, and Sonoma Sky Alpha passed it. All these SOTA models used atleast 1 minute!

Web3Aible's tweet photo. I tested #K2Think by @mbzuai and in one shot, it crushed the most challenging Quest2 in just 7 seconds!
Note only 03, 04mini, gpt-5 (mini-High, High, High-new-system-prompt), Grok 3 mini-High, and Sonoma Sky Alpha passed it. All these SOTA models used atleast 1 minute! https://t.co/2g5iZnu6H9

1

5

0

451

Top Tweets for #k2think

Last Seen Hashtags on Sotwe

Trends for you

Most Popular Users