> be Kimi K3
> drop July 16
> 2.8 trillion parameters
> the biggest open model ever released
> 1 million token context. native multimodal.
> Kimi Delta Attention — 6.3x faster decoding
> priced at a third of Fable 5
> people rush in so fast the GPUs hit their limit in 48 hours
> Moonshot pauses new subscriptions just to keep you running
> the closed labs charge for the frontier
> you're about to be free
> open weights drop July 27
> anyone can run you. anywhere.
> someone already wrote the full guide to building with Kimi
> different game.
> full breakdown below.
before you watch the World Cup final tonight, watch this 30-minute interview with Kimi founder Zhilin Yang
it's the clearest look I've seen at how K2 was actually built and where AI models go next
> why compute scaling still beats every other method
> why the "data flywheel" everyone talks about hasn't formed yet
> the real challenge in multimodality nobody mentions
> why K2 was closer to a base model — the real potential wasn't unlocked yet
this is only part of the interview, translated to English — and everything he hinted at just shipped as Kimi K3, 2.8 trillion parameters, the biggest open model ever
full breakdown below
CHINA CONSIDERS RESTRICTING OVERSEAS ACCESS TO CUTTING-EDGE AI MODELS
China’s Ministry of Commerce has led meetings over the past month with major AI companies, including Alibaba, ByteDance, and https://t.co/YDe0KRldDB, to discuss measures that would restrict overseas access to cutting-edge AI models, including models that have not yet been released.
The discussions reportedly include not only closed-source models but also open-weight models. However, the scope of application is still under debate, and the rules may ultimately apply only to future frontier models.
Officials have also discussed designating the leakage or theft of proprietary AI technologies as a national security crime, with stronger penalties, as well as restricting the types of foreign capital that can invest in Chinese AI startups.
The backdrop is the U.S. move to strengthen export controls on AI models, along with national security concerns over cutting-edge models that could possess advanced cyberattack capabilities.
Chinese authorities are reportedly concerned that advanced U.S. cybersecurity AI models could be used to exploit vulnerabilities in Chinese software.
Since the beginning of this year, China has continued to tighten measures to prevent AI technology from being transferred overseas. Authorities have investigated whether Chinese AI startups that relocated abroad violated export control laws, while also strengthening oversight of overseas transactions involving Chinese investors, technology, data, and national security concerns.
Future regulations could take the form of a tiered framework based on technological capability. Basic open-source AI models may be managed through a filing system, high-performance models may be subject to security reviews, and the most sensitive frontier models may be banned from public release or restricted to use within China.
We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.
Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.
We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.
🧵👇
DROP EVERYTHING
The bible for running LLMs locally is now available online to read for free
Covers what to use on
- Laptop / edge / odd hardware
- Mac-first workflows
- Single RTX GPUs
- 2-4+ NVIDIA / CUDA GPUs
- General production serving
- Long-context / MoE / routing
- NVIDIA max performance
- Cluster orchestration
Software
- llama.cpp
- MLX / MLX-LM
- ExLlamaV2
- ExLlamaV3
- vLLM
- SGLang
- TensorRT-LLM
- NVIDIA Dynamo
You should read this, and if you cannot now then you most definitely wanna bookmark it for later
Local AI FTW
My ALBANIA TOURISM guide... 🇦🇱
Sea / Beach:
• Palasë / Dhërmi / Himarë — basically the Côte d’Azur of Albania. Higher prices across the board, but by far the best beaches, restaurants, and boutique hotels. If budget isn’t a concern, Green Coast is the top choice: gated, clean, modern, and luxurious. I personally bought a duplex with a pool in the new area.
• Ksamil / Sarandë — lower budget compared to the Riviera, beautiful Ionian beaches, lively in summer, and great seafood.
• Durrës / Gjiri i Lalëzit — ideal if you want quick access from Tirana. Gjiri i Lalëzit especially has become one of the more modern coastal areas with resorts, villas, and quieter beaches.
• Rana e Hedhun — one of the most unique spots in the north, where the sand dunes meet the sea. Very calm, underrated, and perfect if you want something less crowded + cheap.
Cities / Nightlife:
• Tirana — a true metropolis and, in my opinion, the best city in the Balkans. Incredible architecture, strong cultural identity, beautiful artificial lakes, one of the best culinary scenes in the Mediterranean region, outstanding bars and rooftops, solid nightlife, great people, and only 30 minutes away from both beaches and mountains.
Historic Destinations:
• Shkodër — historic Venetian influence, stunning lake and castles, strong Catholic culture, and incredibly cheap seafood and drinks.
• Berat — beautiful Ottoman architecture, great food and drinks, museums, and one of the most visually unique cities in the country.
• Gjirokastër — known for its stone architecture, Orthodox heritage, fair prices, and incredible atmosphere.
• Krujë — home of Skanderbeg, filled with history, mountain views, Ottoman-era streets, and one of the most important cultural sites in Albania.
Mountains:
• Kukës — surrounded by mountains and incredible scenery. Not much life, but amazing nature and views.
• Theth — small alpine village, perfect as a base for hikes, rivers, waterfalls, and some of the best landscapes in the Balkans.
hear me out anon. whatever you're using right now, drop it for one night. install hermes agent. set a /goal for that one thing you've been thinking about but never started. go to sleep. just try it once.
Gemma 4 sees a kid and three dogs. Decides what matters. Calls SAM 3.1
Mask and bounding box. Spotlight on subjects. Background blur. Background pixelation.
Four effects. Fully agentic. Two models talking to each other on a MacBook.
No App. No cloud. What would you edit?
MASSIVE
Gemma 4 (31B, Dense), a model that performs on parity w/ Kimi K2.5 (1.1T, MoE)
> 35x SMALLER than Kimi K2.5
Would run on any hardware at home
- RTX 3090 / 4090 / 5090 *
- DGX Spark / Mac Studios
- MacBook Pro (24GB+)
New local SoTA
* Best perf. is when ran on GPUs
A lot of the "private" AI options out there are lying to you. Look under the hood. Use the dev tools on the browser to see what is actually sent. It's plaintext. What they're actually doing is _promising_ they won't look.
We don't do that. See for yourself. We have an inspector tab so you can see the actual traffic data, but you can confirm it in the browser.
https://t.co/qpZVWgArur
Don't trust. Verify.
So Quilibrium just beat ETHs own roadmap, shipping the risc-v zk runtime Vitalik wants. And not only that. Eth plans for 32bit but today 64bit went live.
This enables running full Linux VMs inside Q, along with providing proofs and trustless bridging for Eth and Solana. And using CPUs, no expensive GPU clusters.
There will be some serious Eureka moments when people start to realize what is being built here, and what it implies.
The criticism is tired, and now, accusing of using community as exit liquidity, libelous. We've been shipping updates like crazy, the claims that nothing shipped is beyond horseshit. In the past two months we shipped:
- the mobile beta of Quorum, with a built in wallet, farcaster integration, and mini app support
- QConsole, the management console for Q services
- Qstorage, the S3 compatible file hosting service
- QKMS, the MPC key management service
- QQ, the SQS compatible queue service
- QPing, the SNS compatible notification service
- Klearu, our MPC CPU-powered machine learning library and runtime
- MetaVM, our ZK RISC-V, EVM, and sBPF library
- several Q node software updates, in cooperation with the many diligent and dedicated testers who have been instrumental in seeing things through
Mods banned you from the discord and other venues because you went batshit. But this is exhausting dude, you've followed people talking about real progress happening to try to slander the project, you followed me on both X and Farcaster to repeatedly troll bait. Seek some fucking help.