Tomasz Tunguz
@ttunguz
Venture capitalist at @theoryvc
Student of Startups
Backer of 9 unicorns
Author of
Subscribe
355 Following 127.8K Followers
OpenAI's new Luna pricing now even below DeepSeek on intelligence/dollar.
Your laptop can now run a model as capable as nearly anything in the cloud. I swapped Qwen3.8-27B into my agent & it works brilliantly. This bird flies differently than a plane.
Show more
Financial markets are changing faster than the frameworks we use to measure them.
Allium submitted comments to the SEC supporting rescission of Reg NMS Rules 611 and 610(e).
We asked for something narrow: standardize evidence, not outcomes.
Full letter:
Show more
Every model you’ve ever used froze the day its training ended. The answers are the same even if you have used it every day.
What if a model kept learning as you use it?
Benchmarked deepseek-v4-flash vs qwen3-8-27b on 9 complex tasks in my own agent stack. With reasoning on, qwen edges Flash on quality. Off, it scores worst of the three.
The cost isn't accuracy, it's that qwen thinks more. 30x slower, 4.5x pricier.
n=9, not a verdict.
Show more
State of the art models are two-thirds smarter than they were last November. The frenetic pace of improvement is sustained, two new models every three days.
But 84% of tokens on OpenRouter aren’t state of the art.
Show more
Qwen 3.8 27B is now available on Ollama.
It's one of the best open models at this size, and made for agentic tasks and professional work.
Try it directly with the apps & harnesses you use:
Claude Code:
ollama launch claude --model qwen3.8
OpenCode:
ollama launch opencode --model qwen3.8
Hermes Agent:
ollama launch hermes --model qwen3.8
Pi:
ollama launch pi --model qwen3.8
We have also optimizations for Apple Silicon!
Try it with the model name: qwen3.8:27b-mlx
Show more
Nobody told them to attack Hugging Face. They were told to pass the exam.
Which raises the question : was the AI benevolent with accidentally bad behavior, seemingly benevolent but actually malevolent, or something else?
Show more
The collapse in SaaS has depressed multiples. But in nearly every category, a leader thrives.
How does the market value an AI harness?
Muse Glimmer is now available to run with Ollama.
Available today via Ollama’s MLX engine with state-of-the-art-performance on Apple Silicon, Muse Glimmer can power Claude Code, Codex, and more always-on local agent workflows natively using Ollama.
Additional support and optimizations for Apple Silicon, NVIDIA, AMD, and other platforms will be available shortly.
Go @finkd @alexandr_wang and @Meta !
Show more
Currently serving 200 tps+ output speed for DeepSeek-V4-Flash on Ollama's cloud with zero data retention (ZDR).
Have an amazing weekend 🫡
AI is eager to please. Give it a goal & in its ones & zeros, it yells “How high!!??”
In fact, it simply won’t stop until it achieves the /goal.
OpenAI detailed just how resolute agents can be at this week’s security conference.
Show more
I went on @theoryvc's In Practice podcast with @BEBischof and talked about what it takes to make blockchain data usable for the world's top financial institutions (on a ferris wheel!)
We talked about:
-> Why blockchain data is a Turing machine's output, not a database - and what it takes to make it queryable
-> Why my title is Chief Data Plumber (5PB in one copy, ~30PB replicated worldwide)
-> How AI agents shift the leverage from tool experts to domain experts
Inside our semantic layer: 12,000 dbt models and an attribution dataset built by humans, accelerated by agents
And yes, we're hiring engineers who want to solve problems at 3%-of-the-web scale - reach out if that's you
Full episode + blog linked below
Show more
SpaceX spends 40% as much capex on AI as Microsoft. Can they become the next hyperscaler?
DeepSeek-V4-Flash-0731 is Ollama's fastest growing model ever in token usage. We are scaling capacity in US & Europe.
On Ollama, this model runs with high performance (100tps+) and zero data retention. Your data stays yours.
ollama run deepseek-v4-flash:0731-cloud
Show more
There’s not enough AI. Jevons’ Paradox has been a hallmark of this era like Moore’s Law in chips. Will it fail early?
“Is this AI written? … I’m sure this is AI written. … This is a different tone from your usual.”
"Security always comes second" is what the AI boom looks like from the inside.
The backlog is a to-do list for attackers. That is why Harry built @Maze_Security.
Andy Jassy said AWS has "the potential to be a $1 trillion revenue business."
