NovaSky's Sky-T1-32B-Preview, their fully open-source reasoning model that matches o1-preview on popular reasoning and coding benchmarks — trained under $450!
Blog: novasky-ai.github.io/posts/sky-t1/
Model: huggingface.co/NovaSky-AI/S...
Blog: novasky-ai.github.io/posts/sky-t1/
Model: huggingface.co/NovaSky-AI/S...
January 11, 2025 at 6:44 PM
NovaSky's Sky-T1-32B-Preview, their fully open-source reasoning model that matches o1-preview on popular reasoning and coding benchmarks — trained under $450!
Blog: novasky-ai.github.io/posts/sky-t1/
Model: huggingface.co/NovaSky-AI/S...
Blog: novasky-ai.github.io/posts/sky-t1/
Model: huggingface.co/NovaSky-AI/S...
Open-sourced code and reasoning dataset to train a reasoning model distilled from DeepSeek-R1, provided by Bespoke Labs.
Blog: www.bespokelabs.ai/blog/bespoke...
Code: github.com/NovaSky-AI/S... and github.com/NovaSky-AI/S...
Blog: www.bespokelabs.ai/blog/bespoke...
Code: github.com/NovaSky-AI/S... and github.com/NovaSky-AI/S...
January 23, 2025 at 3:17 AM
Open-sourced code and reasoning dataset to train a reasoning model distilled from DeepSeek-R1, provided by Bespoke Labs.
Blog: www.bespokelabs.ai/blog/bespoke...
Code: github.com/NovaSky-AI/S... and github.com/NovaSky-AI/S...
Blog: www.bespokelabs.ai/blog/bespoke...
Code: github.com/NovaSky-AI/S... and github.com/NovaSky-AI/S...
"Sky-T1-32B-Preview, our reasoning model that performs on par with o1-preview on popular reasoning and coding benchmarks."
That was quick! Is this already the Alpaca moment for reasoning models?
Source: novasky-ai.github.io/posts/sky-t1/
That was quick! Is this already the Alpaca moment for reasoning models?
Source: novasky-ai.github.io/posts/sky-t1/
January 14, 2025 at 12:34 AM
"Sky-T1-32B-Preview, our reasoning model that performs on par with o1-preview on popular reasoning and coding benchmarks."
That was quick! Is this already the Alpaca moment for reasoning models?
Source: novasky-ai.github.io/posts/sky-t1/
That was quick! Is this already the Alpaca moment for reasoning models?
Source: novasky-ai.github.io/posts/sky-t1/
That applies to DeepSeek.
Novasky.
Bluesky.
Linux...
Novasky.
Bluesky.
Linux...
If there's anything you should take away from the DeepSeek moment, it's that open source is an incredible driver of innovation, and we need to push more companies to open source their work
January 29, 2025 at 1:37 AM
That applies to DeepSeek.
Novasky.
Bluesky.
Linux...
Novasky.
Bluesky.
Linux...
NovaSky AI's S*: Test-Time Scaling for Code Generation
S* enables (1) non-reasoning models surpass reasoning models: GPT-4o-mini + S* > o1-preview. (2) open models compete SOTA: R1-Distilled-32B +S* ~= o1 (high).
S* enables (1) non-reasoning models surpass reasoning models: GPT-4o-mini + S* > o1-preview. (2) open models compete SOTA: R1-Distilled-32B +S* ~= o1 (high).
February 22, 2025 at 6:08 PM
NovaSky AI's S*: Test-Time Scaling for Code Generation
S* enables (1) non-reasoning models surpass reasoning models: GPT-4o-mini + S* > o1-preview. (2) open models compete SOTA: R1-Distilled-32B +S* ~= o1 (high).
S* enables (1) non-reasoning models surpass reasoning models: GPT-4o-mini + S* > o1-preview. (2) open models compete SOTA: R1-Distilled-32B +S* ~= o1 (high).
Researchers open source Sky-T1, a ‘reasoning’ AI model that can be trained for less than $450
Researchers open source Sky-T1, a ‘reasoning’ AI model that can be trained for less than $450
So-called reasoning AI models are becoming easier — and cheaper — to develop. On Friday, NovaSky, a team of researchers based out of UC Berkeley’s Sky Computing Lab, released Sky-T1-32B-Preview, a reasoning model that’s competitive with an earlier version…
tcrn.ch
January 11, 2025 at 9:47 PM
Researchers open source Sky-T1, a ‘reasoning’ AI model that can be trained for less than $450
SkyRL tx: An open source project to implement the Thinking Machines' Tinker API, a REST based API for neural network forward/backward passes that unifies inference and training into one common API
Blog: novasky-ai.notion.site/skyrl-tx
Repo: github.com/NovaSky-AI/S...
Blog: novasky-ai.notion.site/skyrl-tx
Repo: github.com/NovaSky-AI/S...
October 7, 2025 at 3:37 AM
SkyRL tx: An open source project to implement the Thinking Machines' Tinker API, a REST based API for neural network forward/backward passes that unifies inference and training into one common API
Blog: novasky-ai.notion.site/skyrl-tx
Repo: github.com/NovaSky-AI/S...
Blog: novasky-ai.notion.site/skyrl-tx
Repo: github.com/NovaSky-AI/S...
The NovaSky team fine-tuned the open-source Qwen2.5-32B-Instruct model. The training lasted for 19 hours using 8 H100 GPUs, costing about $450 (priced according to Lambda Cloud).
January 12, 2025 at 1:43 PM
The NovaSky team fine-tuned the open-source Qwen2.5-32B-Instruct model. The training lasted for 19 hours using 8 H100 GPUs, costing about $450 (priced according to Lambda Cloud).
Data curation, train, eval code, 17K training data: github.com/NovaSky-AI/S...
GitHub - NovaSky-AI/SkyThought: Sky-T1: Train your own O1 preview model within $450
Sky-T1: Train your own O1 preview model within $450 - NovaSky-AI/SkyThought
github.com
January 11, 2025 at 6:44 PM
Data curation, train, eval code, 17K training data: github.com/NovaSky-AI/S...
This is what gemma3n was supposed to help with. Each of the gemma3n models (translation, function, embeddings) can be built on and run on consumer devices.
During the deepseek era, you had projects like NovaSky
novasky-ai.github.io/posts/sky-t1/
The image/sdxl crews are already passing around LoRA
During the deepseek era, you had projects like NovaSky
novasky-ai.github.io/posts/sky-t1/
The image/sdxl crews are already passing around LoRA
January 21, 2026 at 6:34 AM
This is what gemma3n was supposed to help with. Each of the gemma3n models (translation, function, embeddings) can be built on and run on consumer devices.
During the deepseek era, you had projects like NovaSky
novasky-ai.github.io/posts/sky-t1/
The image/sdxl crews are already passing around LoRA
During the deepseek era, you had projects like NovaSky
novasky-ai.github.io/posts/sky-t1/
The image/sdxl crews are already passing around LoRA
'NovaSky, a research lab at University of California, Berkeley, this month released technology it said was on par with a recent model released by OpenAI. The NovaSky scientists built it for $450 by distilling an open-source model from Chinese company Alibaba.' www.wsj.com/tech/ai/why-...
Why ‘Distillation’ Has Become the Scariest Word for AI Companies
DeepSeek’s success learning from bigger AI models raises questions about the billions being spent on the most advanced technology.
www.wsj.com
January 30, 2025 at 6:45 PM
'NovaSky, a research lab at University of California, Berkeley, this month released technology it said was on par with a recent model released by OpenAI. The NovaSky scientists built it for $450 by distilling an open-source model from Chinese company Alibaba.' www.wsj.com/tech/ai/why-...
Novasky drone show on Sat.
November 12, 2024 at 8:52 PM
Novasky drone show on Sat.
On Friday, NovaSky, a team of researchers based out of UC Berkeley’s Sky Computing Lab, released Sky-T1-32B-Preview, a reasoning model that’s competitive with an earlier version of OpenAI’s o1 on a number of key benchmarks.
“Remarkably, Sky-T1-32B-Preview was trained for less than $450,” they said
“Remarkably, Sky-T1-32B-Preview was trained for less than $450,” they said
January 13, 2025 at 1:30 AM
On Friday, NovaSky, a team of researchers based out of UC Berkeley’s Sky Computing Lab, released Sky-T1-32B-Preview, a reasoning model that’s competitive with an earlier version of OpenAI’s o1 on a number of key benchmarks.
“Remarkably, Sky-T1-32B-Preview was trained for less than $450,” they said
“Remarkably, Sky-T1-32B-Preview was trained for less than $450,” they said
Which is where begging for money and locking in contracts became ironic ... But retail investors and consumers aren't going to ever know about similar advances to DeepSeek cooking at Berkeley and in the OSS communities (check out Oumi and Novasky).
February 2, 2025 at 1:22 AM
Which is where begging for money and locking in contracts became ironic ... But retail investors and consumers aren't going to ever know about similar advances to DeepSeek cooking at Berkeley and in the OSS communities (check out Oumi and Novasky).
DAMN! Don't pop on for a few days & WOW! The 🤖!
Good morning All! Halisky, Nova Scotia (NovaSky - sounds like a wine), the rest of Turtle Island! What's on the agendas today? Tunes & housework here b/c I worked second job yesterday when I normally do the majority of odds & ends around the house!
Good morning All! Halisky, Nova Scotia (NovaSky - sounds like a wine), the rest of Turtle Island! What's on the agendas today? Tunes & housework here b/c I worked second job yesterday when I normally do the majority of odds & ends around the house!
January 19, 2025 at 2:27 PM
DAMN! Don't pop on for a few days & WOW! The 🤖!
Good morning All! Halisky, Nova Scotia (NovaSky - sounds like a wine), the rest of Turtle Island! What's on the agendas today? Tunes & housework here b/c I worked second job yesterday when I normally do the majority of odds & ends around the house!
Good morning All! Halisky, Nova Scotia (NovaSky - sounds like a wine), the rest of Turtle Island! What's on the agendas today? Tunes & housework here b/c I worked second job yesterday when I normally do the majority of odds & ends around the house!
a great idea for a province that is shutting down coal plants in the next few years.
#nspoli #novasky #halisky #energysky #energytwitter
www.canarymedia.com/articles/bat...
#nspoli #novasky #halisky #energysky #energytwitter
www.canarymedia.com/articles/bat...
Duke Energy to knock down coal plant and build its biggest battery yet
Large grid batteries are finally coming to North Carolina, the nation’s fourth largest solar power producer.
www.canarymedia.com
December 3, 2024 at 1:43 PM
a great idea for a province that is shutting down coal plants in the next few years.
#nspoli #novasky #halisky #energysky #energytwitter
www.canarymedia.com/articles/bat...
#nspoli #novasky #halisky #energysky #energytwitter
www.canarymedia.com/articles/bat...
Please do. My feed is 💩 today with stupid conspiracy theories
Facts: deepseek methods are published and can be recreated. Novasky is also competitive on smaller hardware. Snowflake AI released a cost/energy cutting product
Kelsey Hightower mentioned the problem today, too
bsky.app/profile/kels...
Facts: deepseek methods are published and can be recreated. Novasky is also competitive on smaller hardware. Snowflake AI released a cost/energy cutting product
Kelsey Hightower mentioned the problem today, too
bsky.app/profile/kels...
Now that DeepSeek is having an impact on the US stock market, I expect a huge negative press cycle designed to discredit their advancement is on the way.
January 27, 2025 at 3:32 PM
Please do. My feed is 💩 today with stupid conspiracy theories
Facts: deepseek methods are published and can be recreated. Novasky is also competitive on smaller hardware. Snowflake AI released a cost/energy cutting product
Kelsey Hightower mentioned the problem today, too
bsky.app/profile/kels...
Facts: deepseek methods are published and can be recreated. Novasky is also competitive on smaller hardware. Snowflake AI released a cost/energy cutting product
Kelsey Hightower mentioned the problem today, too
bsky.app/profile/kels...
They just did it better.
In the meantime no one is even remembering Novasky exists.
novasky-ai.github.io
In the meantime no one is even remembering Novasky exists.
novasky-ai.github.io
January 29, 2025 at 8:45 PM
They just did it better.
In the meantime no one is even remembering Novasky exists.
novasky-ai.github.io
In the meantime no one is even remembering Novasky exists.
novasky-ai.github.io
SkyRL-Agent, a framework for efficient RL agent training.
⚡ 1.55× faster async rollout dispatch
🛠 Lightweight tool + task integration
🔄 Backend-agnostic (SkyRL-train / VeRL / Tinker)
Repo: github.com/NovaSky-AI/S...
Paper: arxiv.org/abs/2511.16108
⚡ 1.55× faster async rollout dispatch
🛠 Lightweight tool + task integration
🔄 Backend-agnostic (SkyRL-train / VeRL / Tinker)
Repo: github.com/NovaSky-AI/S...
Paper: arxiv.org/abs/2511.16108
November 27, 2025 at 4:16 PM
SkyRL-Agent, a framework for efficient RL agent training.
⚡ 1.55× faster async rollout dispatch
🛠 Lightweight tool + task integration
🔄 Backend-agnostic (SkyRL-train / VeRL / Tinker)
Repo: github.com/NovaSky-AI/S...
Paper: arxiv.org/abs/2511.16108
⚡ 1.55× faster async rollout dispatch
🛠 Lightweight tool + task integration
🔄 Backend-agnostic (SkyRL-train / VeRL / Tinker)
Repo: github.com/NovaSky-AI/S...
Paper: arxiv.org/abs/2511.16108