achieve ambition with intentionality, intensity, integrity & insanity. affiliations: - @smol_ai - @cognition - @temporalio - @aidotengineer - @latentspacepod
san francisco / singapore
Joined April 2009
- Tweets78.7K
- Following4K
- Followers208K
- Likes136K
Pinned Tweet
💁♂️ How to Play Long Term Games:
Systems > Goals
Discipline > Motivation
Trust > Distrust
Principles > Tactics
Writing > Reading
Vulnerability > Confidence
North Stars > Low Hanging Fruit
Trends > News
Habits > Sprints
Questions > Answers
Problems > Solutions
People > Projects
Gemini #1 on AAv5???
this is making me open up @antigravity for the first time in a long time. i’m getting an early review, will report back soon.
if google actually pulls this off it is officially comeback season with Meta, Grok and now GDM being cool again.
@saranormous talked about model neurodiversity last year. we have roughly 2x more frontier labs than when she was last on the aie stage. absolutely bizarre counter example to the bitter lesson.
Google’s Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5, and second only to Claude Opus 5.5 on the Artificial Analysis Coding Agent Index
Gemini 4 Carbon scores 59 on Intelligence Index v5, driven by leading performance in complex reasoning, knowledge, and scientific tasks. It also has large gains in coding and knowledge work, particularly when compared to other Gemini models, but it is not leading in all these areas.
As with launch of Gemini 4 Argon, Gemini 4 Carbon is being rolled out to selected users and is not yet publicly available. Google is offering a 50% launch discount, with no confirmed end date. All our charts and figures use standard pricing for both Gemini 4 Carbon and Argon.
Key results for Gemini 4 Carbon with high reasoning:
➤ Gemini 4 Carbon leads the Intelligence Index v5, with the #1 score in 4 of the 9 Index evaluations: HLE-Diamond, Terminal-Bench-Science, AA-Omniscience, AutomationBench-AA. Carbon makes substantial gains elsewhere but is not leading in coding and knowledge work
➤ Behind only Claude Opus 5.5 on coding agents: With Antigravity CLI, Gemini 4 Carbon (high) scores 59 only 3 points behind Claude Code with Claude Opus 5.5 (max, 62) on the Artificial Analysis Coding Agent Index. Carbon ranks second on both Terminal-Bench 4.0 (64%) and FrontierCode v1.1 (51%), behind Claude Opus 5.5 (max) at 56%. Google has historically underperformed in this area and so this improvement stands out
➤ Improved agentic knowledge work, but not yet leading: On AA-Briefcase, Gemini 4 Carbon (high) scores 1609 Elo, a 126-point improvement over Gemini 4 Argon (high, 1483), but still behind models including Grok 4.7 (xhigh, 1638), Muse Wonder (max, 1659) and Kimi K3.1 Pro (1678). The largest gain comes from Presentation, up 293 Elo to 1598, alongside a 120-point improvement in Analytical Quality to 1671
➤ Lower Cost per Task than Claude Opus 5.5 for comparable overall intelligence: Gemini 4 Carbon costs $7.76 per Intelligence Index task, ~23% less than Claude Opus 5.5 (max, $10.04) but ~1.2x GPT-6 Astra (max, $6.25). It uses 123k output tokens per task, 29% fewer than Claude Opus 5.5 (max, 173k). At the same $4/$20 standard pricing, its Cost per Task is in line with Gemini 4 Argon ($7.68), while scoring 9 points higher on the Intelligence Index
➤ Leading AA-Omniscience v1.1 score, with low hallucination: Gemini 4 Carbon scores 54 on our recently updated AA-Omniscience v1.1, 6 points ahead of Claude Opus 5.5 (max, 48) and 8 points ahead of Gemini 4 Argon (46). Accuracy (% correct) rises 10 points over Argon to 58%, while its hallucination rate rises from 4% to 11%. This is still the second lowest of any model, behind only Argon, compared with 50% for both Claude Opus 5.5 (max) and GPT-6 Astra (max)
Key model details:
➤ Context Window: 1M tokens
➤ Multimodality: Text, image, video, and speech input, with text output
➤ Pricing: $4/$20 per 1M input/output tokens, the same as Gemini 4 Argon’s standard pricing, with a 50% discount. Cached input tokens receive a 95% discount ($0.20 per 1M)
Congratulations to the @GoogleDeepMind team!
swyx retweeted
introducing prism, the world’s first speech-to-speech model you can’t tell from a human in conversation
prism has:
> 300ms response time, 4x faster than OpenAI GPT-Live
> outperforms every other s2s model on intelligence (with NO reasoning)
> the first s2s model customizable through reinforcement learning
we’re a team of 35 with $35M in funding, and we’re unlocking s2s model adoption faster than openai, google, and x.
the first 100 people to sign up here get 1 month free: rime.ai/prism?utm_campaign=p…
at spacex ipo everyone was posting their cursor bag fumble.
kid stuff.
my turn.
i estimate this was a $500m bag fumble lol. and very probably wouldve been on claude code early (bc i've been a typescript cli maintainer). fml.
will you buy $ANT stock at a $2T valuation?
14%Yes, significant % of NW
23%Yes, "average" investment
38%No, its for retail chumps
25%Undecided/Maybe
6,659 votes • 4 daysanyway here's my soundcloud teapawt.org/latentspacepod/status/…
How Anthropic ships Claude with Claude
latent.space/p/opus55/
Opus 5.5 is the #1 LLM in the world right now. We go behind the scenes of the launch, with @katelyn_lesse and @angjiang, heads of Eng and Product behind the platform used to build everything from the first party harnesses like Code/Tag/Duet and by >50% of Anthropic's top customers!
swyx retweeted
My favourite event I ever hosted was a karaoke and trivia night in New York ❤️
We brought the community members of Latent Space together to sing our hearts out - WHO KNEW THEY’D HAVE SUCH BARS!!!
Thank you to my partner in crime @KaylinWardley ❤️
Until the next time :)
yeah you ignored a @mntruell email in 2024 but did you pitch something like Claude Code in 2019 and then.... not do it for 7 years gist.github.com/swyxio/3af1e…
swyx retweeted
An interesting idea on calibration of LLMs from @swyx when they were discussing Jev and decision models in @latentspacepod:
> There’s one element of calibration which is a rare topic that I don’t think people even knew about or understood. We covered it with, our conversation with Clémentine Fourrier of Hugging Face, and she used to run the evals, at Hugging Face, which is basically the idea that, models are attuned to give you the most likely next token. But, they’re gonna lie to you when you ask them, “How confident are you?” Because they’re just gonna give you the most likely next answer instead of, like, actually, like, no, let’s calibrate. Like, I am actually fifty percent sure, or I am twenty percent sure, and, like, let’s try to calibrate that. I would say, like, if anything, I think that actually that’s pretty easy to generate synthetic data around because you can sort of see the truth and then synthetically generate a bunch of answers, have Jev classify it, and then compare with ground truth. That would be my reverse engineering of this.
It was very nice to hear this because we've been exploring something very similar in our recent work on ProbabilityBench: creating synthetic data where we have access to the exact ground-truth probabilities to evaluate decision models and other LLMs. arxiv.org/abs/2610.17016
Calibration metrics are becoming popular for evaluating decision models, but they don't directly measure how close predicted probabilities are to the ground truth probabilities. A model can be perfectly calibrated while still producing inaccurate probability estimates for individual instances. So I also feel evaluating (and/or training) on synthetically generated data with ground truth probabilities might be a good way to go. Previously we had to rely on toy experiments with Gaussian distributions where we can analytically derive the ground truth probabilities (which is not suitable for testing LLMs), or had to use benchmarks with crude soft labels, e.g., CIFAR-10H, Fashion-MNIST-H, and ChaosNLI.
Jev already uses synthetic data, and they may be doing something similar internally. While we don't know the details, some experiments shared in X suggest there's still room for improvement: when asked about a fair six-sided die, Jev assigned an average probability of ~83% to its chosen outcome (instead of 17%), and ~92% for a fair coin outcome (instead of 50%). This highlights how challenging getting accurate probabilities can be, even in these very simple settings. (see details in our blog: ishida-lab.github.io/blog_pr… )
It would be exciting to see more research in this direction!
Recursive Language Models: Claude Code, Agent Swarms, Big Research Bets, & the $40M AI Problem Solver latent.space/p/rlm
MIT’s @a1zhang explains why Claude Code, Codex, and Pi are basically the same, how RLMs use code, context offloading, and recursive subagents to generalize across tasks, why one expert can sometimes replace a trillion-token brute-force search, what OpenAI’s 10,000-agent, 130B-output-token experiment reveals about the future of language models, and why academia’s biggest advantage is the freedom to take weird, ambitious research bets.
swyx retweeted
One of the more under-appreciated details in Anthropic's S-1:
Nearly half of Anthropic's revenue now flows through cloud marketplaces - 48% in the first half of 2026, up from 11% in 2023. That's roughly $7.8B of revenue in six months sold through Bedrock, Vertex and Azure Marketplace instead of direct.
Which means the hyperscalers get paid twice! They sell Anthropic the compute - Anthropic has committed hundreds of billions of dollars to AWS, Google and Microsoft - and then they take a fee on roughly half of the revenue distributed through their marketplaces.
Also on gross margin...those marketplace fees don't sit in cost of revenue. They're in sales, marketing & partnerships, so they never touch the headline 85% gross margin everyone is quoting today. Fees paid to channel partners alone rose ~$2.7B year over year in H1 2026.
Hyperscalers feel like they're kind of the winner here all around though reading through this.
swyx retweeted
going to be sharing my WoW forever shaman build at the AIE Code closing keynote
We are pleased to announce that @trq212 will be returning to AIE to share what's now and what's next!
aie.codes will be the last flagship AIE of the year, gathering the top 1000 agentic engineers in the world by invite only.
Apply here: ai.engineer/code/2026/apply
As usual all keynotes livestreamed and talks posted free on our youtube!
swyx retweeted
Diogo Almeida and the TypeSafe AI team are bringing Jev to AIE CODE.
System One models and what people are building with them.
November 10 to 12, San Francisco.
Apply to attend: ai.engineer/code
ALT AI Engineer Code Summit card showing Diogo Almeida in his bright pink jacket. Co-founder and CEO of TypeSafe AI. November 10 to 12, San Francisco. ai.engineer/code.
swyx retweeted
Your coding agent's verbosity is a code review problem.
Ben Holmes at AIE NY on scoring redundant tests and noisy explanations.
Scaling without Slop at AIE CODE.
November 10 to 12, San Francisco.
Apply to attend: ai.engineer/code
swyx retweeted
glm 5.3 max reasoning effort top 2 on first problem, top 1 on the second
ranking:
Problem 1 (target 2025)
GPT-6 Astra: 8.53s
GLM-5.3: 10.87s
Scott Wu: ~13.2s
Opus 5.5: 13.46s
Gemini 4 Argon: 19.87s
Muse Wonder: 64.65s
Grok 4.7: 94.02s
Problem 2 (target 6843)
GLM-5.3: 4.50s
Opus 5.5: 10.17s
Gemini 4 Argon: 13.48s
GPT-6 Astra: 19.74s
Scott Wu: ~20.8s
Grok 4.7: 49.53s
Muse Wonder: 50.70s
we are hiring writers, reviewers and reporters for the new @latentspacepod editorial team (hit up @ricmac and @adlinzainal) and i’m gonna put a version of this in the welcome package for everyone we hire
part time and oneoff guest posts welcome docs.google.com/forms/d/e/1F…
swyx retweeted
Honestly, I was surprised @WSJ wanted to do a profile on me. 😂
I’m not a cool 20-year-old college dropout. I’m a pretty boring 40-year-old trying to put everything I’ve learned throughout my career to work building a new market infrastructure for compute.
And if I succeed, the headline won’t be true for very long.
Wall Street will have the data, transparency, and tools to understand, price, and manage compute risk better than I ever could.
And that would be the real success.
Carmen Li hopes to make AI computing power an asset class like oil or gas that investors can bet on—or against. on.wsj.com/4yEEXGL
