team @noat_future || ex HFT trader || 🇸🇪 🇺🇸
Los Angeles, CA
Joined January 2021
- Tweets12.4K
- Following5.9K
- Followers2.4K
- Likes101K
benchmarcᵐᶠᵉʳ retweeted
Liquid Inference is serving 10B tokens/day and growing 200% week over week.
Most of the traffic is agents and long chats. Across 470k charged jobs, 88.7% of input tokens were cache reads.
Here’s how our matching algorithm prices offers with cache hit rate included:
In multi-turn and agentic workloads, the biggest cost and latency wins come from prefix/KV cache reuse. Each turn resends the full conversation, so if a provider still has that prefix in its cache it can bill those tokens at its cache-read rate, often a small fraction of the normal input rate.
Naive load balancing or pure price auctions scatter a session across different providers or replicas, forcing full prefill on every turn and destroying the savings.
Our matching algorithm ranks offers on the expected charge inclusive of cache hit rate rather than the full list price:
1. We measure how much of each workspace’s input repeats from one turn to the next (7-day window, at least 50 jobs).
2. We multiply that share by the measured cache hit rate of each deployment.
3. We price those tokens at the offer’s cache-read rate and subtract the expected saving from the offer’s ranking charge.
4. The deployment that served the previous turn keeps a 5% margin, so a cheaper offer has to win by more than 5% to pull a warm session away. In other words, for the right price we will abandon the cache and choose the cheaper alternative.
The credit only changes the order of the offers. A wrong estimate cannot make a job cost more than its price cap. Every receipt shows the cache share we expected, the share that was realized, and the credit in dollars.
That’s how we continue to improve on cache hit rate and provide the best economics for our clients’ sessions.
benchmarcᵐᶠᵉʳ retweeted
Payward, the parent company of Kraken, announced its intention to be the first CFTC-regulated deployer on HIP-3* markets. Payward will use Hyperliquid as the infrastructure layer to deploy perpetual futures markets for US clients. Subject to regulatory approval, Bitnomial, Payward’s CFTC-regulated DCM and DCO, would act as the HIP-3* deployer: creating, owning, and administering the markets and clearing and settling the contracts. If approved, NinjaTrader Clearing, Payward’s CFTC-registered FCM, would carry client accounts.
HIP-3 will support optional deployer configuration for permissioned markets in a future network upgrade. This set of additional deployer functionality is collectively called HIP-3*. At its core, HIP-3* activation involves an onchain allowlist managed by the deployer or its sub-deployers. The functionality is strictly additive to HIP-3; its use is optional and up to the deployer. Existing markets remain unchanged. The initial release is available on testnet, where specifications are preliminary and subject to change based on feedback.
Hyperliquid is a neutral infrastructure layer with the goal of supporting large parts of the financial system. HIP-3* gives deployers additional functionality to operate deployments consistent with the requirements applicable to them. As with HIP-3, HIP-3* deployers are independent operators who use Hyperliquid as the onchain infrastructure layer for their own markets, while retaining control and responsibility over their deployment.
benchmarcᵐᶠᵉʳ retweeted
There is an absolute barrage of inference providers popping up that get to $1M/mo extremely quickly.
This is the rough economics of inference providers:
Even if you rent on-demand B300s at $8/GPU-hr, many open source models (GLM 5.3, Kimi K3, DeepSeek 4.1 Flash / Pro) can get you >$10-20/GPU-hr at reasonable throughputs (100tok/s). DeepInfer from DeepSeek directly shows they can get $38/GPU-hr on off-peak hours. Every chip / model has a different throughput / interactivity curve, but there is a lot of arbitrage between even high rental spot prices and token prices. New chips (Vera Rubins) also change this meaningfully. Your ability to be making gross margin relies on your utilization and how good your stack is.
Gateways like OpenRouter, Vercel and more help solve that problem by aggregating demand to you. May inference providers discount their tokens to get longer term commits too. I've even heard stories of selling back Chinese model tokens to Chinese labs!
One thorn in the economics of these players is that many Chinese open source model labs are now asking for rev share if you're at a certain scale, espp for pre-release access to their models for day 1 support. That further pushes the threshold to be in the money.
The barrier to entry seems low and falling. There are even stories of smaller ones who can "vibe code" a way to stand up OSS models on a cluster they rent. And as a result, we're seeing 10s of such companies pop up almost every week and immediately skyrocketing revenue at low-ish gross margin and raising rounds on the back of that.
Unsurprisingly, because the actual service provided seems commodity, this eventually drives up compute prices and drives down token prices. In theory, it would seem like neoclouds who have the lowest cost of compute and themselves have started inference businesses would win the race to the bottom (Crusoe Coreweave Nebius and GMI all have inference businesses), with larger players (Modal Fireworks Baseten) being second because they are well capitalized to make bigger compute commits cheaply and have demand (high utilization), but the reality is the pie is growing so quickly that even upstarts are absolutely crushing it and growing aggressively (inference dot net, DeepInfra, Wafer, Relace, Morph).
Strategies vary across the board: neoclouds want to serve inference, inference players want to be neoclouds and do higher margin reserved instances / post training, and some want to go heterogeneous hardware or creatively flexible compute constructs.
The risks are: do open source models continue to be competitive? Will their creators rent seek more? Will their shape change compute needs? How do different chips change economics? Who gets compute in a supply crunch?
Below is a aggressive sketch of how gross margin scales with utilization for various models / chip sets.
For now, there are many inference providers and they’re all ripping!
benchmarcᵐᶠᵉʳ retweeted
Don't miss posters today at #ESMO26: Salomon Tendler will present data showing that DLL3 PET identifies neuroendocrine differentiation and tumor heterogeneity in advanced prostate cancer. @Tendler_S @MSKCancerCenter
📅 October 23 | 3:15 p.m. CEST
ALT Promotional graphic for poster presentation featuring Dr. Salomon Tendler.
benchmarcᵐᶠᵉʳ retweeted
FULL INTERVIEW: Emmy winner @jayleno joins the show to discuss comedy, supercars, the state of Hollywood, and more:
0:00 - Cars
11:00 - The disappearance of the middle class
16:00 - Comedy & politics
17:30 - The future of late-night television
19:00 - Why Hollywood has become so expensive
30:00 - Turning down business opportunities
32:00 - Why comedians should stick to comedy
39:00 - Retirement and staying relevant
47:00 - Advice to AI doomers
benchmarcᵐᶠᵉʳ retweeted
.@jayleno believes low self-esteem is the key to success.
"If you don't think you're the smartest person in the room, just shut up and pay attention."
"That's what I did when I got on the Tonight Show."
"Give me the best lighting guy, give me the best director, and please tell me when I suck."
"I have friends of mine in the business and nobody tells them the truth."
"They'll say, 'Oh, you're good. You're really good.'"
"No, you weren't. Whatever you're doing, change it because it's not working."
benchmarcᵐᶠᵉʳ retweeted
dear GEOTUS @realDonaldTrump I have a high verbal IQ and am very adept at harassing journalists. I was captain of the debate team. I am not female, blonde or particularly attractive but I believe I can compensate for that in the form of witty repartee. I offer you my services as press secretary entirely free of charge.
benchmarcᵐᶠᵉʳ retweeted
Founders: Are you on text with your business customers?
Met with a startup founder who had just been texting his largest customer (which informed him over text that they were rolling out the product more broadly).
Texting with your best accounts is a sign of strong, trusted relationships. In fact, if you're NOT texting, it means that the relationship is not strong enough.
Get on text!
benchmarcᵐᶠᵉʳ retweeted
These are the states where you can walk into a gas station and buy a THC beverage.
Federal hemp rules shut most of it down December 11.
benchmarcᵐᶠᵉʳ retweeted
Meta gave its AI away for free for years. This week they finally built one good enough to charge for, and they won't sell it. There's no price. You literally cannot buy Meta's best model at any cost.
Sounds backwards until you see what they're doing.
Every other AI lab makes money the same way. You pay per word the model reads and writes. OpenAI charges for tokens, Anthropic charges for tokens, Google charges for tokens. The price sheet is the whole business.
Meta just opted out of that business. Their smaller new model, Muse Wonder, you can buy like normal, $4 in and $20 out per million tokens, basically Claude pricing. The bigger one, Muse Wonder Full, the one that just tied Claude on coding benchmarks and sits 2 points behind Opus 5.5 overall, has no price sheet at all. The only place it exists is inside Muse Code.
And honestly, if you're Zuck, why would you sell it?
Meta already has what every other lab is burning billions to get. Distribution. Muse did 2.5 million downloads in 13 days and hit #1 on the App Store. Amazon got nervous enough to block it. WhatsApp puts this thing one tap away from 3 billion people.
Once you own the pipes, selling tokens leaks your advantage. Anyone can buy your API output and use it to train their own model, and your frontier intelligence ends up as a line item inside a competitor's product. Keep it locked inside your own apps and the only way anyone touches Meta's best AI is by becoming a Meta user.
Meta spent two years lecturing the industry about open AI, Llama weights free for anyone to download. The lectures stopped the same month their models got good.
Meta has released Muse Wonder and Muse Wonder Full, achieving Artificial Analysis Intelligence Index v4.3 scores of 52 and 56 at max reasoning effort - each a material gain over Muse Spark 1.3 (max) at 48, with a release gap of only 6 weeks
Muse Wonder is available for users on Meta’s first-party API, while Muse Wonder Full is currently only available for Muse Code users. Meta notes that each model is designed for personal and professional agent use
Congratulations to @AIatMeta, @finkd, and @alexandr_wang on the release!
Key Takeaways:
➤ Both models exhibit a material jump in overall Intelligence Index score. Muse Wonder Full (max) improves on Muse Spark 1.3 (max) across all ten Intelligence Index evaluations, gaining 8 points to 56. The largest gains were seen in agentic evaluations: Terminal-Bench 4.0 (+24 points, 33% to 57%), GDPval-AA v2.1 (+176 Elo, 1681 to 1857), and AA-Briefcase (+148 Elo, 1581 to 1729). Muse Wonder (max) gains 4 Intelligence Index points and lands at 52, with notable improvements vs. Muse Spark 1.3 (max) in Terminal-Bench 4.0 (33% to 52%) and AutomationBench-AA (58% to 68%)
➤ Muse Wonder Full (max) places in the top three on six of the ten Intelligence Index evaluations, and is particularly strong in agentic knowledge work and scientific reasoning. In agentic knowledge work, it narrowly leads on GDP.pdf (33%), ranks #2 on GDPval-AA v2.1 (1857 Elo), and #3 on both AA-Briefcase (1729 Elo) and AutomationBench-AA (71%). In scientific reasoning, it tops the CritPt leaderboard at 33% and ranks #3 on SciCode with 63%. This builds on the relative strengths of Muse Spark 1.3 (max), which also showed particular ability in these two domains
➤ Both models use fewer output tokens than Muse Spark 1.3 (max). Muse Wonder (max) and Muse Wonder Full (max) each use around 54k output tokens per Intelligence Index task, down ~10% from Muse Spark 1.3 (max, 60k). This usage is lower than similarly intelligent models Gemini 4 Argon (high, 62k) and Claude Fable 5.1 (max w/ fallback, 78k), but higher than GPT-6.1 Sol (max, 38k) and GPT-6 Astra (max, 27k). Muse Wonder Full (max) uses 16% fewer input tokens per Intelligence Index task than Muse Wonder (max) (6.44M vs 7.69M), meaning it is both more intelligent and more token-efficient
➤ Cost per Task for Muse Wonder (max) lands higher than most similarly intelligent models. Despite efficient output token use, premium API pricing ($4 per million input tokens, $20 per million output tokens) and high input token use (7.69M per task) push Muse Wonder (max)’s cost to $4.52 per Intelligence Index task. This is lower than Claude Fable 5.1 (max w/ fallback, $7.63), Claude Opus 5.5 (max w/ fallback, $5.98), and Claude Sonnet 5.5 (max w/ fallback, $5.46), but higher than GPT-6 Astra (max, $3.26), Gemini 4 Argon (high, $1.99), GPT-6.1 Sol (max, $0.72), and its predecessor Muse Spark 1.3 (max, $1.60). Per-token pricing is not currently available for Muse Wonder Full, so no Cost per Task figure can be estimated at this stage
➤ Both models gain significantly in Coding Agent performance. Muse Wonder Full achieves 68 on the Artificial Analysis Coding Agents Index at max effort in Muse Code, alongside Claude Code with Sonnet 5.5 (max, 68) and Opus 5.5 (max, 66). Muse Wonder (max) scores 63. These represent material gains over Muse Code with Muse Spark 1.3 (max), which scores 54
Other details for the two new models:
➤ Context windows: 1M tokens each, unchanged from Muse Spark 1.3
➤ Pricing: for Muse Wonder, $4 input / $0.20 cached input / $20 output per 1M tokens. Pricing per token is not yet announced for Muse Wonder Full
➤ Input modalities: text, image, video, and audio
➤ Output modalities: text
➤ Availability: Muse Wonder is available on Meta's first-party API and Muse Code at launch, while Muse Wonder Full is only available to Muse Code users
benchmarcᵐᶠᵉʳ retweeted
JUST IN: In-N-Out's refusal to offer delivery has reportedly sparked a "burger black market" in California, with people paying up to $50 extra to get burgers delivered.
benchmarcᵐᶠᵉʳ retweeted
Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.
benchmarcᵐᶠᵉʳ retweeted
A federal judge agreed to halt Altria Group Inc.’s challenge against the US Food and Drug Administration’s policy for evaluating premarket tobacco applications, as the agency seeks to revise the current framework. news.bloomberglaw.com/health…
benchmarcᵐᶠᵉʳ retweeted
If I were stranded on an island and could live with only one person, I’d pick Grok.
benchmarcᵐᶠᵉʳ retweeted
All definitions of "addictive" boil down to:
Addiction = Dependence + Harm
Caffeine, for example, is dependence-forming but not "addictive."
Nicotine itself does not cause harm. In fact, it has some health benefits...
benchmarcᵐᶠᵉʳ retweeted
CNN published a deeply misleading story and headline suggesting that my diplomatic work in the Middle East somehow benefited investments in Israeli defense companies.
But CNN’s own reporting says otherwise. Buried behind a paywall, it acknowledges: “CNN found no indication that Kushner’s actions as a diplomat directly affected the investments reviewed.”
Phoenix is not a defense company. It is one of Israel’s largest publicly traded financial institutions, highly regulated and managing approximately $220 billion across thousands of investments.
Think of Phoenix as the Israeli equivalent of Vanguard or Fidelity. It manages the retirement savings of millions of Israelis. The savers don’t choose the individual investments, and neither do I. I have no role in deciding what Phoenix buys or sells. It has performed very well for Affinity and our investors, up more than five times since our initial investment.
Affinity is an SEC-registered investment firm. Our investment in Phoenix is public and dates to before President Trump was re-elected and before I returned to public service. Affinity does not own the defense companies highlighted in CNN’s story.
Those facts make for a less provocative headline, but a much more accurate story.
Misleading stories will not distract me from the work President Trump asked me to do: help end wars, bring people together and pursue peace. That work is too important.
benchmarcᵐᶠᵉʳ retweeted
Smoking Is Bad for You. Nicotine Isn't—but Politicians Keep Trying To Ban It.
reason.com/2026/10/01/smokin…
