Conway is a AI, cryptography, and product lab that deeply cares about distribution. We are behind @underdogai
👾
Joined November 2025
- Tweets80
- Following3
- Followers12.8K
- Likes212
Conway Research retweeted
Today we’re releasing Underdog Bark 3.0, the #1 frontier speech model
It's a tiny 163 MB download as accurate as NVIDIA’s Parakeet (15x bigger). It beats OpenAI's Whisper large-v3 (57x bigger)
It's fast- transcribes 1 min of audio in <0.25s on a Macbook. On recorded meetings (AMI), Bark 3.0 gets 7.82% of words wrong. Parakeet gets 9.42%. OpenAI's Whisper large-v3 gets 13.63% error rate
Bark runs on both your Laptop & Phone. It’s free and open source under CC-BY 4.0
Try private on-device meeting notes & dictation, powered by Bark, in Underdog today: underdog.ai/bark
Conway Research retweeted
Happy to partner with @NVIDIARTXSpark
Saluki runs fast on RTX Spark🐕
Bringing more capacities and fast running doggos to the hood!
Love seeing what @UnderdogAI is bringing to RTX Spark. Their Saluki 27B model brings fast, private personal AI to your laptop in under 8GB of memory. 👏
Conway Research retweeted
assembled a sick team of MTS
now adding one to the growth team
you’ll be working with 160 billboards, the coolest designers, the creators behind the biggest memes and stores in prime locations. Good pay & equity & lots of fun & getting roasted while making history
you don’t need to be a ghostwriter. We all speak our minds, online or in person. you don’t need to invent a brand story. you just need to amplify the truth. be creative, be happy, and handle some human coordination with the legacy agencies we haven’t automated away yet
send to your fren. If your fren gets in, you get cash or bones in the dogpark :)
join us so we can do the dog dance tgt
underdog.ai/team?v=2
Conway Research retweeted
Thanks for the invite! cant wait to bring magic to NVIDIA RTX Spark! 🙏@NaderLikeLadder @ChrisMlachowsky
We're partnering with @NVIDIA to bring frontier intelligence to @NVIDIARTXSpark laptops
Underdog's model and chip codesigned inference engine optimizes CUDA Megakernels to bring Datacenter speeds, on a laptop
Same weights & hardware. Every update makes Underdog faster with better software data efficiency
Underdog Saluki 27B is on par with Claude Opus 4.6 and fits in under 8 GB
Underdog: your private personal AI, now on NVIDIA devices underdog.ai/nvidia
Conway Research retweeted
it’s a pleasure to work with the @NVIDIA team for the past few weeks
excited to bring frontier intelligence at datacenter speeds with CUDA Megakernels & model/chip codesigned inference engines
stay tuned! we are going to make magic with this insanely powerful laptop!
shoutout @NaderLikeLadder @gerdelgado & the amazing NVIDIA team & leadership
We're partnering with @NVIDIA to bring frontier intelligence to @NVIDIARTXSpark laptops
Underdog's model and chip codesigned inference engine optimizes CUDA Megakernels to bring Datacenter speeds, on a laptop
Same weights & hardware. Every update makes Underdog faster with better software data efficiency
Underdog Saluki 27B is on par with Claude Opus 4.6 and fits in under 8 GB
Underdog: your private personal AI, now on NVIDIA devices underdog.ai/nvidia
Conway Research retweeted
300,000+ downloads in the first 7 days! Underdog Saluki 27B is Trending on Huggingface 🤗
Users are saying that it's the #1 model that's <8 gb (soon <6gb). We 🩵 the open source local AI community!
Try Saluki today in Underdog - Your Private Personal AI
huggingface.co/ConwayResearc…
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
Introducing Underdog x Stripe
Shop with Underdog. It finds the best option, saving you money and time. In a local browser, it gets a one-time Link card for the exact amount, and waits for your approval
The first on-device Agent Payment in Underdog: Your Private Personal AI
Conway Research retweeted
free, on-device image generation & editing in Underdog
Introducing Underdog Images: on-device Image Gen
Generate any image, or drop in a photo and say what to change. It all runs on your device. Your photos never leave your computer
Made with Qwen-Image Flash from @NVIDIAAI and Qwen-Image-2.1-Turbo from @Alibaba_Qwen
Conway Research retweeted
Introducing Underdog Videos: on-device video generation
Free & fast generation from frontier video models
Make any videos, bring any picture to life, and add sound- fully local. Nothing leaves your device
Conway Research retweeted
Introducing Underdog Images: on-device Image Gen
Generate any image, or drop in a photo and say what to change. It all runs on your device. Your photos never leave your computer
Made with Qwen-Image Flash from @NVIDIAAI and Qwen-Image-2.1-Turbo from @Alibaba_Qwen
Conway Research retweeted
A 27B model shrinks from 54GB to 7.89GB,
that fits under 8GB and beats the full-size version at tool calling.
Saluki keeps 96% of Qwen 3.8 27B performance at roughly 1/7th the size and runs entirely on your Mac locally.
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
wait, a 2 bit file just beat the full bf16 qwen 3.8 27b at tool calling 👀
underdog squeezed qwen 3.8 27b from 54gb down to 7.89gb and on their bench it passed 88 of 120 tool calls against 84 for the full bf16, and 42 parallel calls against 35, same harness for both.
tool calling is what agents run on, so if 2 bit keeps it a 27b agent fits on a 12gb card with room for context
one thing i want to see with my own eyes, their chart beats bonsai 2's smallest file, 5.95gb at 1.75 bits, 88 to 70. bonsai's own 2 bit file is 7.21gb, that's the fair fight at the same size
both go on the same 12gb card next, i'll post what holds
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Try Saluki 27B today in Underdog underdog.ai/saluki
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
A dense 27B model in 7.89 GB, running at ~22 tok/s on my DGX Spark with 262k context. And 87 GB of memory still free. 🤯
Saluki is pretty wild.
If you want to try it yourself:
github.com/dedene/Underdog-S…
→ docker compose up
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
For friends with 8-16gb of vram!
New best model for you
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
Just downloaded and activated Underdog. Onboarding is just WOW.
Unsubscribed so many unwanted newsletters in few seconds. 💪
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
Today we're releasing Underdog Saluki 27B
Based on Qwen 3.8 27B, Saluki is almost 7x smaller than its full-precision counterpart while it keeps 96% of its benchmark performance and beats Qwen at tool calling
It's the best model <8GB that runs in Underdog
Saluki is available today under Apache 2.0. Try today in Underdog underdog.ai/saluki
Conway Research retweeted
Local AI is accelerating, better be ready, trust me!
Introducing Underdog, your Private Personal AI on devices you already own
Today we're announcing our backing from @a16z @khoslaventures @HummingbirdVC Anthology (@AnthropicAI @MenloVentures) @patrickc @naval @rauchg @polynoamial @Thom_Wolf @Mascobot @tszzl @OfficialLoganK and other top AI leaders
Our mission is to provide free, capable, reliable and private AI to billions of people.
Underdog’s Law: today’s frontier intelligence reaches your devices in six months.
We’re starting with fast, capable AI that runs on consumer hardware. By co-designing models and inference engines, we’re pushing the frontier of capability, speed, power and data efficiency.
Our research also spans agentic commerce, confidential inference, and how AI will reshape the internet economy.
Privacy and capability no longer needs to be a tradeoff. If you believe in this future, join us. Time to build.
Conway Research retweeted
Sigil Wen, backed by a Silicon Valley's who-who, has built an on-device AI assistant that promises to be free, fully private, and capable for everyday tasks.
spr.ly/6011BGFgL7
Conway Research retweeted
Want an invite to underdog.ai ?
