dean

[Addy Osmani @addyosmani]

https://t.co/hIe0UX7z6T

[Bilawal Sidhu @bilawalsidhu]

God's eye view 24-hour replay of Operation Epic Fury.

The Iran strikes kicked off and I set an AI agent swarm loose to record every OSINT signal I could find before the caches cleared. Built a full 4D reconstruction in WorldView.

I can scrub through minute by minute and watch the whole thing unfold on a 3D globe:

> Airspace clearing over Tehran
> Ground strike coordinates locking in
> Severe GPS interference blinding the region
> EO and SAR satellites making passes over the strike zone
> No-fly zones locking down 9 countries
> Shipping fleets scrambling at the Strait of Hormuz

It's pretty amazing how complete of a picture you can build without "proprietary data fusion" -- one dev with public signals and a love for computer graphics and geospatial intelligence.

Thank you for all the love on my last post. Dropping WorldView in April. This my friends is just the beginning.

[Woosuk Kwon @woosuk_k]

Today, we're proud to announce @inferact, a startup founded by creators and core maintainers of @vllm_project, the most popular open-source LLM inference engine.

Our mission is to grow vLLM as the world's AI inference engine and accelerate AI progress by making inference cheaper and faster.

The Challenge
Inference is not solved. It's getting harder.

Models grow larger. New architectures proliferate: mixture-of-experts, multimodal, agentic. Every breakthrough demands new infrastructure. Meanwhile, hardware fragments: more accelerators, more programming models, and more combinations to optimize.

The capability gap between models and the systems that serve them is widening. Left this way, the most capable models remain bottlenecked and with full scope of their capabilities accessible only to those who can build custom infrastructure. Close the gap, and we unlock new possibilities.

And the problem is growing. Inference is shifting from a fraction of compute to the majority: test-time compute, RL training loops, synthetic data.

We see a future where serving AI becomes effortless.

Today, deploying a frontier model at scale requires a dedicated infrastructure team. Tomorrow, it should be as simple as spinning up a serverless database. The complexity doesn't disappear; it gets absorbed into the infrastructure we're building.

Why Us
vLLM sits at the intersection of models and hardware: a position that took years to build.

When model vendors ship new architectures, they work with us to ensure day-zero support. When hardware vendors develop new silicon, they integrate with vLLM. When teams deploy at scale, they run vLLM, from frontier labs to hyperscalers to startups serving millions of users. Today, vLLM supports 500+ model architectures, runs on 200+ accelerator types, and powers inference at global scale. This ecosystem, built with 2,000+ contributors, is our foundation.

We've been stewards of this engine since its first commit. We know it inside out. We deployed it at frontier scale—in research and in production.

Open Source
vLLM was built in the open. That's not changing.

Inferact exists to supercharge vLLM adoption. The optimizations we develop flow back to the community. We plan to push vLLM's performance further, deepen support for emerging model architectures, and expand coverage across frontier hardware. The AI industry needs inference infrastructure that isn't locked behind proprietary walls.

Join Us
Through the open source community, we are fortunate to work with some of the best people we know. For @inferact, we're hiring engineers and researchers to work at the frontier of inference, where models meet hardware at scale. Come build with us.

We're fortunate to be supported by investors who share our vision, including @a16z and @lightspeedvp who led our $150M seed, as well as @sequoia, @AltimeterCap, @Redpoint, @ZhenFund, The House Fund, @strikervp, @LaudeVentures, and @databricks.

- @woosuk_k, @simon_mo_, @KaichaoYou, @rogerw0108, @istoica05 and the rest of the founding team

[Paul Graham @paulg]

I bet few people there know it, but YC wouldn't exist without UIUC. My father went there from England on a Fullbright, and came back determined to move to America.

[Guillermo Rauch @rauchg]

There are no limits anymore. Anyone can do anything. The only limiting factors are agency and ambition.

Never has a college degree, work experience, network, even the accumulation of knowledge been worth less.

You can just ship things.

[dean @deankepler]

@Teknium lol

[OpenRouter @OpenRouter]

We collaborated with @a16z to publish the **State of AI** - an empirical report on how LLMs have been used on OpenRouter.

After analyzing more than 100 trillion tokens across hundreds of models and 3+ million users (excluding 3rd party) from the last year, we have a lot of insights to share.

[dean @deankepler]

@durov interesting

[dean @deankepler]

@evanjconrad congrats

[dean @deankepler]

@tensorblock_aoi @AnthropicAI ok we shipped it, y’all have fun!!

[dean @deankepler]

shipped and live. excited to see what people build with opus 4.5 on forge.

[TensorBlock @tensorblock_aoi]

Claude Opus 4.5 (@AnthropicAI ) is now supported on Forge.

Anthropic’s newest model brings stronger coding and agentic behavior with better SWE bench and tool-use performance.

On Forge, you can use Opus 4.5 through the same single API as every other supported provider.

[TensorBlock @tensorblock_aoi]

Claude Opus 4.5 (@AnthropicAI ) is now supported on Forge.

Anthropic’s newest model brings stronger coding and agentic behavior with better SWE bench and tool-use performance.

On Forge, you can use Opus 4.5 through the same single API as every other supported provider.

[Armin Ronacher ⇌ @mitsuhiko]

Followup to yesterday's post: I'm starting to think as agents and LLM APIs of being a state synchronization problem and that we might look into what the local first folks are doing. Dumped my thoughts here: https://t.co/M66kxMgsyE

[dean @deankepler]

https://t.co/qX4oLrF67O

[dean @deankepler]

read the gemini 3 model card. it is clearly not a fine tuned 2.5, but a newly trained sparse moe with a new backbone.

on top of gemini 2.5’s strong rl post training and parallel thinking, this new backbone makes gemini 3 a real step up.

last six months of gemini:
- strong rl based post training
- parallel thinking as a default behavior
- a new sparse moe backbone
- open benchmarks that actually help the field, like imo bench

last six months of openai:
- adaptive thinking mainly about saving enterprise costs and treating users as second class
- agent mode basically forgotten
- a proactive assistant that looks like an ads surface
- credit expiration that keeps pressure on user spending
- sam’s never ending hype on top

this is the difference between a company building for the field and a company optimizing the funnel.

https://t.co/qgMgCDU0vi

[TensorBlock @tensorblock_aoi]

Gemini 3.0 is now live on Forge.

You can call Google’s newest model through the same unified TensorBlock API you already use across providers.

One interface, many models. Switch providers without changing your application logic.

Try it here → https://t.co/lNxtpBg9D3

[dean @deankepler]

@Teknium lol

[dean @deankepler]

That is why the same alt shows up on top gainers today and top losers tomorrow.

The arb vs. MM game is getting more interesting lol

[dean @deankepler]

Altcoin tickers are slots.

When the room is packed and bankroll is thick, people bid up the machines so everyone can play.
When cash is thin, the machines get sold for scraps.

Buying micro caps = value investing in slots.
But in a cash-light room, you’re just taking someone’s crate of machines off their hands.

In 2022, the floor mcap for a @binance perp was ~$25M.
Pre-Oct 11 it was ~$10M.
Now ~$8M.
Next probably ~$6M, idk..

[dean @deankepler]

2025.11.7
I’m writing this to talk about altcoins. Most people come into crypto to get rich fast, not to hold BTC and grind out a steady return.

First, this space is a true negative-sum market. Most altcoins do not create any real value, so whatever wealth appears comes from outside capital flowing in.

BTC keeps printing new highs, while the ceiling for altcoins keeps getting lower. Look at the pure retail markets. $PEOPLE, $ORDI, and 币安人生 on Binance are all classic retail coins with no real lead sponsor. In the 2021 bull market $PEOPLE was only a viral meme, not even a new sector narrative, and it still ran to around 1.2 billion. $ORDI, as a new Bitcoin meme narrative, topped at roughly 2 billion. 币安人生, a newer Chinese narrative, is already getting heavily sold around 400 million. Every cycle the cap gets lower, because the easy retail money is gone and people have become much more cautious.

And the reason is obvious. The naive believers in alts have mostly been wiped out. The ones who remain understand that the fundamental value of most altcoins is essentially zero.

So where did altcoin wealth come from in the past? It came from outside money brought in by majors like BTC and ￥$ETH. Early retail accumulated BTC. When Wall Street arrived, retail was attracted by the volatility in alts and sold BTC to chase them. Even if they made money at first, they usually lost it back when the bear market arrived. Cycle after cycle, the dividends earned from BTC by retail slowly moved upward to more advanced players.

A clearer way to see this is through three characters. A is the naive buyer who bought BTC before 2018. B entered in 2021 and is an average participant. C entered in 2023 and is the sharp one. A later sold BTC and chased alts, and because A lacked skill, A was wiped out during the 2021 to 2022 cycle. B was a bit smarter and made money in the first round, but during the 2023 to 2024 PVP, B’s wealth moved to C.

So the wealth that remains in the hands of today’s players came from A stacking BTC early, Wall Street buying A’s BTC, and then a series of alt PVP and perp PVP transfers that moved wealth from A to B to C.

You can find A and B and C everywhere in this space. This is a snapshot of the era. C looks like late but very capable entrants such as @0xAnteater, @HsakaTrades, @awawat,@SalsaTekila, @NachoTrades, @smartestmoney , @alphawifhat, @Gold_Cryptoz and @ganymede_0x.

A typical example of B would be @machibigbrother. Most of his money was made by extracting value from retail during the 2021 liquidity wave. Even though we have many mutual friends, the truth is that $CREAM benefited him massively during the 2021 cycle. But as more teams and more projects appeared, it became harder for him to make money from launching new tokens. If you look at his secondary trades in $FT, $XPL, $PUMP, and $Blur NFTs, most of them have been heavy losses. Especially $PUMP, where he bought aggressively and still ended up selling the bottom. A typical retail outcome. The reason he made so much money back then was simply because BTC brought in a huge amount of outside capital that spilled onto him.

Last year in the @TreeNewsFeed Discord I said you do not need to hold $PEPE anymore. someone eventually understood what I meant. Most of the time you can simply hold USDT or BTC. When a hot narrative appears, trade it. When the heat fades, rotate back to USDT or BTC...

Altcoins have zero value. I genuinely do not understand why anyone would hold large altcoins long term. They do not rise much, they crash harder, and over time they underperform BTC and SP500.

Most of my own wealth came from picking up bodies during black swan events and from early news trading. In 2019 the bodies were $EOS, $BCH, and $LTC. In 2021 they were $DOT and $LINK. In 2025 they were $DOGE and $SUI. Large altcoins are especially likely to blow up during black swans because forced sellers appear at the same time.

Think about it. $UNI and $ATOM and $ORDI are all just different stories wrapped around the same mechanics. Even $XPL is the same kind of story, only aimed at different people. And whales are not always smart. Someone literally bought thirty million dollars worth of $XPL at the top.🤣

Crypto is a real casino. The only assets with lasting value are BTC, platform tokens with real revenue, certain DeFi primitives, and pump formats that monetize gambling flow, and even that revenue ultimately comes from speculation.

Face the reality. Crypto is speculation. When a narrative appears, it becomes a PVP arena. When the narrative dies, you either hold USDT or you hold the majors.

This space is easier to make money in than before, and also harder. Easier because the market is larger and sharp players have more opportunities. Harder because if you rely only on clean trading with no shortcuts or shady tricks, you face people who survived multiple cycles of destruction. In late 2019, an on chain PVP pot might have been one hundred thousand dollars and you could move it easily. Now you cannot even enter that size without leaving a trace.

Even without trading, you can make money as a KOL. A large group of low effort accounts shit post every day and run ads. Anyone with even a bit of influence can get monthly payments from exchanges like @Binance, @OKX, or @Bybit_Official etc.

But if you want to earn purely from trading, you are facing people who crawled out of piles of bones and kept going. It feels like a kid walking into a pro MMA gym with nothing, surrounded by fighters who have taken real hits and learned how to finish a fight...