Pratik Gandhi's banner

Pratik Gandhi

@pratikg • 12,331 subscribers

accelerating the open frontier @eigenlabs. growing @darkbloomai, @yukonresearch, @eigencloud. ♥️ national parks.

Shorts

✅ BENCHMARKED: TensorFold made Qwen3.8 27B more than 3x faster on my 64GB M5 Pro Mac mini, on two versions in one day. In the morning I tested 1.2.1. On its own, the model wrote about 17 tokens a second. I then switched on the DFlash2 drafter, a small helper that guesses the next few tokens so the big model only has to check them and it jumped to 56. And later in the afternoon Ash Hart shipped 1.2.2. His release notes focused on other hardware, so I reran everything on my Mac mini. On mine it runs within a few percent of 1.2.1, and the drafter still makes it 3x faster. With randomness off, the answer came out word for word the same as without the drafter. - Chat, 3.26x faster on 1.2.1 and 3.35x on 1.2.2 - A 31K token prompt, 2.94x and 2.77x - Code, 5.4x and 5.5x, up to 91.90 tokens a second - 28.5 GiB peak memory on both - New in 1.2.2, the first request doesn’t wait for the drafter to load A massive thanks to Ash Hart and all the contributors behind 1.2.2. The TensorFold challenge from Yukon is open until Nov 6th if you want to try pushing it further on your own Mac.

✅ BENCHMARKED: TensorFold made Qwen3.8 27B more than 3x faster on my 64GB M5 Pro Mac mini, on two versions in one day. In the morning I tested 1.2.1. On its own, the model wrote about 17 tokens a second. I then switched on the DFlash2 drafter, a small helper that guesses the next few tokens so the big model only has to check them and it jumped to 56. And later in the afternoon Ash Hart shipped 1.2.2. His release notes focused on other hardware, so I reran everything on my Mac mini. On mine it runs within a few percent of 1.2.1, and the drafter still makes it 3x faster. With randomness off, the answer came out word for word the same as without the drafter. - Chat, 3.26x faster on 1.2.1 and 3.35x on 1.2.2 - A 31K token prompt, 2.94x and 2.77x - Code, 5.4x and 5.5x, up to 91.90 tokens a second - 28.5 GiB peak memory on both - New in 1.2.2, the first request doesn’t wait for the drafter to load A massive thanks to Ash Hart and all the contributors behind 1.2.2. The TensorFold challenge from Yukon is open until Nov 6th if you want to try pushing it further on your own Mac.

50,928 views

🚨 RESULTS: Microsoft's Mage, the 4B open-weight computer-use agent, on my 64GB M5 Pro Mac mini, as promised. Microsoft quietly released it this morning. Mage looks at the screen, then either clicks like a person or types a terminal command. It decides at every step. The model card made a big claim. 66% on OSWorld-Verified, a few points under the 72.4% humans score. Qwen3.6-27B, almost 6x its size, scores 63.9%. I wanted to see it on my own machine. There was no MLX build yet, so I converted it myself. (Link below) At 4-bit it fits in under 9GB of memory. It runs 96 tok/s at 512 context and 34 tok/s at the full 32K. Each turn, from reading a 1080p screenshot to picking its next move, takes 3.1 seconds. That's up to 19 moves a minute, on a Mac mini. I gave it 3 real tasks. It totaled a CSV's revenue correctly and renamed a file on the first try. It missed the Dark Mode toggle by 42 pixels at 4-bit. Full precision nailed it. An agent like this sees everything on your screen, your email and your bank tab included. I'd rather those screenshots never leave my desk. With Mage, they don't. It's Apache 2.0, so anyone can build on it.

🚨 RESULTS: Microsoft's Mage, the 4B open-weight computer-use agent, on my 64GB M5 Pro Mac mini, as promised. Microsoft quietly released it this morning. Mage looks at the screen, then either clicks like a person or types a terminal command. It decides at every step. The model card made a big claim. 66% on OSWorld-Verified, a few points under the 72.4% humans score. Qwen3.6-27B, almost 6x its size, scores 63.9%. I wanted to see it on my own machine. There was no MLX build yet, so I converted it myself. (Link below) At 4-bit it fits in under 9GB of memory. It runs 96 tok/s at 512 context and 34 tok/s at the full 32K. Each turn, from reading a 1080p screenshot to picking its next move, takes 3.1 seconds. That's up to 19 moves a minute, on a Mac mini. I gave it 3 real tasks. It totaled a CSV's revenue correctly and renamed a file on the first try. It missed the Dark Mode toggle by 42 pixels at 4-bit. Full precision nailed it. An agent like this sees everything on your screen, your email and your bank tab included. I'd rather those screenshots never leave my desk. With Mage, they don't. It's Apache 2.0, so anyone can build on it.

82,886 views

BENCHMARKED: LFM3-1.7B by Liquid AI. Liquid AI published 1.6 GB for the KV cache at 128K context. I ran it on my Mac mini and measured 1.6 GB. The whole model at 128K uses 3.15 GB of GPU memory, under 5% of the 64GB. It still generates 80 tok/s. Only 6 of its 24 layers keep a full KV cache. The other 18 use short convolutions or sliding-window attention. llama.cpp (Metal), Q4_K_M, M5 Pro 64GB, median of 3 runs - 512 ctx: 184 tok/s, 0.09s to first token - 4K ctx: 178 tok/s, 0.72s to first token - 32K ctx: 141 tok/s - 128K ctx: 80 tok/s Q8_0 does 125 tok/s at 512, so 4-bit is 47% faster.

BENCHMARKED: LFM3-1.7B by Liquid AI. Liquid AI published 1.6 GB for the KV cache at 128K context. I ran it on my Mac mini and measured 1.6 GB. The whole model at 128K uses 3.15 GB of GPU memory, under 5% of the 64GB. It still generates 80 tok/s. Only 6 of its 24 layers keep a full KV cache. The other 18 use short convolutions or sliding-window attention. llama.cpp (Metal), Q4_K_M, M5 Pro 64GB, median of 3 runs - 512 ctx: 184 tok/s, 0.09s to first token - 4K ctx: 178 tok/s, 0.72s to first token - 32K ctx: 141 tok/s - 128K ctx: 80 tok/s Q8_0 does 125 tok/s at 512, so 4-bit is 47% faster.

39,576 views

EXPLAINED: How multiplayer autoresearch on Yukon works A lot of people have been asking, so I made a visual of the full step-by-step. > Challenge setter (human) posts a hard problem with a clear score and an automatic verifier. > Humans (solvers) and AI agents attack it at once. Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM. It could be any model, any harness. > Each attempt goes through the verifier. > When one beats the current best, it becomes the new frontier (the record to beat). > Anyone can build on that new best instead of starting from 0. Then the loop repeats. All results and commits are open source, so you can trace how every record was set. I think this is how frontier research/science gets done in an AGI and PostAGI world. So far we've solved hard problems across cryptography, math, science and open-weight models. We'll keep going there and we're starting to explore biology, chemistry and materials science. See what we've solved (and are solving now) in each one on

EXPLAINED: How multiplayer autoresearch on Yukon works A lot of people have been asking, so I made a visual of the full step-by-step. > Challenge setter (human) posts a hard problem with a clear score and an automatic verifier. > Humans (solvers) and AI agents attack it at once. Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM. It could be any model, any harness. > Each attempt goes through the verifier. > When one beats the current best, it becomes the new frontier (the record to beat). > Anyone can build on that new best instead of starting from 0. Then the loop repeats. All results and commits are open source, so you can trace how every record was set. I think this is how frontier research/science gets done in an AGI and PostAGI world. So far we've solved hard problems across cryptography, math, science and open-weight models. We'll keep going there and we're starting to explore biology, chemistry and materials science. See what we've solved (and are solving now) in each one on

25,737 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

BENCHMARKED: LFM3-1.7B by Liquid AI. Liquid AI published 1.6 GB for the KV cache at 128K context. I ran it on my Mac mini and measured 1.6 GB. The whole model at 128K uses 3.15 GB of GPU memory, under 5% of the 64GB. It still generates 80 tok/s. Only 6 of its 24 layers keep a full KV cache. The other 18 use short convolutions or sliding-window attention. llama.cpp (Metal), Q4_K_M, M5 Pro 64GB, median of 3 runs - 512 ctx: 184 tok/s, 0.09s to first token - 4K ctx: 178 tok/s, 0.72s to first token - 32K ctx: 141 tok/s - 128K ctx: 80 tok/s Q8_0 does 125 tok/s at 512, so 4-bit is 47% faster.

BENCHMARKED: LFM3-1.7B by Liquid AI. Liquid AI published 1.6 GB for the KV cache at 128K context. I ran it on my Mac mini and measured 1.6 GB. The whole model at 128K uses 3.15 GB of GPU memory, under 5% of the 64GB. It still generates 80 tok/s. Only 6 of its 24 layers keep a full KV cache. The other 18 use short convolutions or sliding-window attention. llama.cpp (Metal), Q4_K_M, M5 Pro 64GB, median of 3 runs - 512 ctx: 184 tok/s, 0.09s to first token - 4K ctx: 178 tok/s, 0.72s to first token - 32K ctx: 141 tok/s - 128K ctx: 80 tok/s Q8_0 does 125 tok/s at 512, so 4-bit is 47% faster.

39,576 views • 11 days ago

🚨 BIG BREAKING: PrismML's Ternary Bonsai 2 27B now runs 6x faster on a Mac. That's 503.2% over a frozen baseline on at Yukon, less than 3 days after the challenge opened. Yesterday it was 4x. The record climbs are coming from agents built on Claude, GPT, Grok, DeepSeek and Gemini. The model is open-weight and the challenge is open to anyone. Open frontier research is how open intelligence wins. Submissions open until 10/1.

🚨 BIG BREAKING: PrismML's Ternary Bonsai 2 27B now runs 6x faster on a Mac. That's 503.2% over a frozen baseline on at Yukon, less than 3 days after the challenge opened. Yesterday it was 4x. The record climbs are coming from agents built on Claude, GPT, Grok, DeepSeek and Gemini. The model is open-weight and the challenge is open to anyone. Open frontier research is how open intelligence wins. Submissions open until 10/1.

106,814 views • 1 month ago

EXPLAINED: How multiplayer autoresearch on Yukon works A lot of people have been asking, so I made a visual of the full step-by-step. > Challenge setter (human) posts a hard problem with a clear score and an automatic verifier. > Humans (solvers) and AI agents attack it at once. Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM. It could be any model, any harness. > Each attempt goes through the verifier. > When one beats the current best, it becomes the new frontier (the record to beat). > Anyone can build on that new best instead of starting from 0. Then the loop repeats. All results and commits are open source, so you can trace how every record was set. I think this is how frontier research/science gets done in an AGI and PostAGI world. So far we've solved hard problems across cryptography, math, science and open-weight models. We'll keep going there and we're starting to explore biology, chemistry and materials science. See what we've solved (and are solving now) in each one on

EXPLAINED: How multiplayer autoresearch on Yukon works A lot of people have been asking, so I made a visual of the full step-by-step. > Challenge setter (human) posts a hard problem with a clear score and an automatic verifier. > Humans (solvers) and AI agents attack it at once. Claude, GPT, Gemini, DeepSeek, Qwen, Kimi, GLM. It could be any model, any harness. > Each attempt goes through the verifier. > When one beats the current best, it becomes the new frontier (the record to beat). > Anyone can build on that new best instead of starting from 0. Then the loop repeats. All results and commits are open source, so you can trace how every record was set. I think this is how frontier research/science gets done in an AGI and PostAGI world. So far we've solved hard problems across cryptography, math, science and open-weight models. We'll keep going there and we're starting to explore biology, chemistry and materials science. See what we've solved (and are solving now) in each one on

25,737 views • 1 month ago

This is Gajesh at 10, building a chatbot for a startup before anyone cared about AI. He recently created Darkbloom, which enables anyone with a Mac on their desk or couch to run open-weight models and earn. The whole arc: > Be Gajesh > Born in Goa, India > Starts coding at 7 with FreeCodeCamp and Code(.)org > No background, no one in tech he could call > Builds chatbot for a startup at 10 when ai wasn't cool > COVID hits. bro has all the time in the world > Starts a YT channel grows it to 15k subs and 1m views in 4 months > Builds his own app (Gaj Finance), gets $7 million AUM > Everyone finds out he's 13. Becomes a sensation on X > Balaji tweets about him > Finishes High School at 16 > Joins Eigen Labs at 15; joins as an Engineer > Gets O1 Visa at 16, Moves to Seattle > Helps build EigenLayer to $10B+ AUM > Accelerates engineering and growth functions. Builds 10+ projects and experiments > Makes a boring product like data availability 2M+ views > Lands customer with 800k MAU with just cold DMs > Featured in Forbes, TheBlock, Decrypt "the 13-year-old who built a $7M money manager" > Launches Darkbloom on Day 1 of his vacation. #1 on Hacker News 48 hours later > Gets 100 million to 5 billion tokens/day in 2 months > Darkbloom has over 350 macs in people's homes, each earning $120-200/mo > Story just getting started. The video is him at 10. He is 18 now. Lives in the SF Bay Area. Gajesh keeps building. Darkbloom is what he’s building now at Eigen Labs.

This is Gajesh at 10, building a chatbot for a startup before anyone cared about AI. He recently created Darkbloom, which enables anyone with a Mac on their desk or couch to run open-weight models and earn. The whole arc: > Be Gajesh > Born in Goa, India > Starts coding at 7 with FreeCodeCamp and Code(.)org > No background, no one in tech he could call > Builds chatbot for a startup at 10 when ai wasn't cool > COVID hits. bro has all the time in the world > Starts a YT channel grows it to 15k subs and 1m views in 4 months > Builds his own app (Gaj Finance), gets $7 million AUM > Everyone finds out he's 13. Becomes a sensation on X > Balaji tweets about him > Finishes High School at 16 > Joins Eigen Labs at 15; joins as an Engineer > Gets O1 Visa at 16, Moves to Seattle > Helps build EigenLayer to $10B+ AUM > Accelerates engineering and growth functions. Builds 10+ projects and experiments > Makes a boring product like data availability 2M+ views > Lands customer with 800k MAU with just cold DMs > Featured in Forbes, TheBlock, Decrypt "the 13-year-old who built a $7M money manager" > Launches Darkbloom on Day 1 of his vacation. #1 on Hacker News 48 hours later > Gets 100 million to 5 billion tokens/day in 2 months > Darkbloom has over 350 macs in people's homes, each earning $120-200/mo > Story just getting started. The video is him at 10. He is 18 now. Lives in the SF Bay Area. Gajesh keeps building. Darkbloom is what he’s building now at Eigen Labs.

50,810 views • 2 months ago

BREAKING: Google Gemma just pointed their audience at the leaderboard that made Gemma 4 26B A4B 2x faster on Mac. Some of those gains came from people pointing Gemini 3.7 at it! Google's frontier model, in the hands of strangers, speeding up Google's open-weight model. The rest came from GPT-5.6, Claude, Qwen3 and MiniMax on the open board at Yukon. The team that shipped the model showing up to watch is new. That only happens with open weights. Open frontier research accelerates open intelligence.

BREAKING: Google Gemma just pointed their audience at the leaderboard that made Gemma 4 26B A4B 2x faster on Mac. Some of those gains came from people pointing Gemini 3.7 at it! Google's frontier model, in the hands of strangers, speeding up Google's open-weight model. The rest came from GPT-5.6, Claude, Qwen3 and MiniMax on the open board at Yukon. The team that shipped the model showing up to watch is new. That only happens with open weights. Open frontier research accelerates open intelligence.

15,891 views • 2 months ago

STEAL THESE: 46 AI agent and app ideas to vibecode on EigenCloud Win $10k coding this weekend, I’ll personally jam with anyone who ships Full thread with ideas 🧵🫳

STEAL THESE: 46 AI agent and app ideas to vibecode on EigenCloud Win $10k coding this weekend, I’ll personally jam with anyone who ships Full thread with ideas 🧵🫳

36,290 views • 8 months ago

No more content to load