ModelsPublished Top storyAll news from that day
Reflection AI's Laskin says open models will take most token demand
On the No Priors podcast, published on YouTube on 9 October, Misha Laskin of Reflection AI gave his account of Beam's training run and his forecast for open models.
Check our sources · 11 facts from 1 source
Video: No Priors: AI, Machine Learning, Tech, & Startups on YouTube
Key points
- Laskin says Beam's reinforcement-learning stage used a little over 10,000 GB300s for four weeks (12:47).
- He says Beam tends to be three to four times more efficient than models of the same capability class, his own claim (20:27).
- He says gateway token use flipped from about 70/30 closed to 70/30 open over roughly six months (25:27), and expects most token demand to go to open models (26:03).
What happened
Misha Laskin, introduced by the host as co-founder and chief executive of Reflection AI, spoke on the No Priors podcast, which published the interview on YouTube on 9 October. Asked for an update, he said Reflection is now at around 300 people.
Of Beam's training, he said the run used 6,000 GB300 chips for a few weeks, and that with infrastructure efficiencies it could now be done in about 12 days or less. He put the reinforcement-learning stage at a little over 10,000 GB300s for four weeks.
He said Beam tends to be three to four times more efficient than models of the same capability class. On the market, he said token use on model gateways has flipped from roughly 70 per cent closed to 70 per cent open, and that he expects most token demand to go to open models.
Asked what it costs to catch up to the frontier, he gave a range of single-digit billions of dollars. On safety, he said closed labs have a few hundred safety researchers and argued that more eyeballs make vulnerabilities shallow.
What it means for you
Our viewLaskin's figures are his own account on a podcast, and he cited no published figure for the shift in gateway token use. He speaks for a company that is building its own model, Beam, so the efficiency comparison and the forecast are best read as his view, not as measurements.
For a team choosing models, three to four times better efficiency matters only on your own tasks. Test it: if you get access to Beam, run the same set of real tasks on it and on your current model, and compare the cost of each completed task, not the list price.
It adds no new facts.
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against this source.
1 Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin
Open the source archived copy-
The No Priors episode "Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin" was published on YouTube on 9 October 2026, and its description says the guest is Reflection AI's co-founder and CEO.
citeBeam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin (title); uploadDate 2026-10-09T10:28:41-07:00; description: ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo and Elad Gil to discuss the launch of Beam
-
The host introduced the guest at the start of the interview (00:42): "Today we're joined by Misha Alaskan, the co-founder and CEO of Reflection AI", the automatic captions' rendering of Misha Laskin (00:42).
cite00:42 Today we're joined by Misha Alaskan, the / 00:44 co-founder and CEO of Reflection AI.
-
Asked for an update on where Reflection is, Misha Laskin said: "we're now at around 300 people" (01:48). The question came from the host and he answered it.
cite01:48 to build one of these things like a a true rocket ship project and we're now / 01:50 at around 300 people
-
Laskin said of Beam: "to train beam which is a 500 billion parameter model total 23B active it was 6,000 uh GB300's we ran it I think for for a few weeks" (12:25).
cite12:25 learning and to train beam which is a / 12:28 500 billion parameter model total 23B / 12:31 active it was 6,000 uh GB300's we ran it / 12:37 I think for for a few weeks
-
Laskin said that "with infrastructure efficiencies we can do it in about 12 days maybe less" (12:39), speaking of the same training run.
cite12:37 ... but now you / 12:39 know like with infrastructure / 12:41 efficiencies we can do it in about 12 / 12:43 days maybe less
-
Laskin said: "reinforcement learning was a little over 10,000 GB300s for four weeks" (12:47).
cite12:47 efficiencies Um, reinforcement learning / 12:50 was / 12:52 a little over 10,000 GB300s for four / 12:54 weeks. So, actually there are more flops / 12:55 spent on reinforcement learning.
-
Asked about Beam's reasoning efficiency, Laskin said: "beam tends to be three to four times more efficient than uh models of the same you know capability class" (20:27).
cite20:27 in one. So beam tends to be three to / 20:31 four times more efficient than uh models / 20:33 of the same you know capability class
-
Laskin said that maybe six months earlier token use on any gateway was majority closed, minority open, and "it's flipped almost exactly from 7030 closed open to 7030 open close" (25:27), the automatic captions' rendering of 70/30 closed to open and 70/30 open to closed. He named the gateways in the captions as "open router or versell" (25:24), that is OpenRouter or Vercel. It is his account; he cited no published figure.
Toned down to what the source says
cite25:19 maybe 6 months ago, it was majority / 25:21 closed uh minority open when you go to / 25:24 any gateway like open router or versell / 25:27 and it's flipped almost exactly from / 25:29 7030 closed open to 7030 open close.
-
Laskin said: "we'll see most token demand going to open" (26:03).
cite26:03 So I think that um we'll see most token / 26:06 demand going to open.
-
Asked how much capital it takes to catch up to the frontier, Laskin said: "I think that now or let's say even six months ago is probably order billion so billions of dollars single digit billions" (09:18).
cite09:18 >> I think that now or let's say even six / 09:21 months ago is probably order billion so / 09:24 billions of dollars single digit / 09:25 billions.
-
On safety, Laskin said: "we have a few hundred safety researchers within closed labs that understand how these things work" (47:11), and "I have the belief that with enough eyeballs um most security and safety vulnerabilities become shallow as well" (47:01). The host had asked about the safety criticism of open models (44:44) and he was answering.
cite47:01 >> and I have the belief that with enough / 47:03 eyeballs um most security and safety / 47:06 vulnerabilities become shallow as well. / 47:09 >> So like the state of the world today is / 47:11 that we have a few hundred safety / 47:13 researchers within closed labs that / 47:15 understand how these things work
Every sentence above is checked against this source. Each fact has its own address you can cite.
Source 1 · No Priors (YouTube) · 9 Oct 2026
Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin
The No Priors episode "Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin" was published on YouTube on 9 October 2026, and its description says the guest is Reflection AI's co-founder and CEO.
Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin (title); uploadDate 2026-10-09T10:28:41-07:00; description: ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo and Elad Gil to discuss the launch of Beam
The host introduced the guest at the start of the interview (00:42): "Today we're joined by Misha Alaskan, the co-founder and CEO of Reflection AI", the automatic captions' rendering of Misha Laskin (00:42).
00:42 Today we're joined by Misha Alaskan, the / 00:44 co-founder and CEO of Reflection AI.
Asked for an update on where Reflection is, Misha Laskin said: "we're now at around 300 people" (01:48). The question came from the host and he answered it.
01:48 to build one of these things like a a true rocket ship project and we're now / 01:50 at around 300 people
Show all 11 factsShow fewer facts
Laskin said of Beam: "to train beam which is a 500 billion parameter model total 23B active it was 6,000 uh GB300's we ran it I think for for a few weeks" (12:25).
12:25 learning and to train beam which is a / 12:28 500 billion parameter model total 23B / 12:31 active it was 6,000 uh GB300's we ran it / 12:37 I think for for a few weeks
Laskin said that "with infrastructure efficiencies we can do it in about 12 days maybe less" (12:39), speaking of the same training run.
12:37 ... but now you / 12:39 know like with infrastructure / 12:41 efficiencies we can do it in about 12 / 12:43 days maybe less
Laskin said: "reinforcement learning was a little over 10,000 GB300s for four weeks" (12:47).
12:47 efficiencies Um, reinforcement learning / 12:50 was / 12:52 a little over 10,000 GB300s for four / 12:54 weeks. So, actually there are more flops / 12:55 spent on reinforcement learning.
Asked about Beam's reasoning efficiency, Laskin said: "beam tends to be three to four times more efficient than uh models of the same you know capability class" (20:27).
20:27 in one. So beam tends to be three to / 20:31 four times more efficient than uh models / 20:33 of the same you know capability class
Laskin said that maybe six months earlier token use on any gateway was majority closed, minority open, and "it's flipped almost exactly from 7030 closed open to 7030 open close" (25:27), the automatic captions' rendering of 70/30 closed to open and 70/30 open to closed. He named the gateways in the captions as "open router or versell" (25:24), that is OpenRouter or Vercel. It is his account; he cited no published figure.
Toned down to what the source says25:19 maybe 6 months ago, it was majority / 25:21 closed uh minority open when you go to / 25:24 any gateway like open router or versell / 25:27 and it's flipped almost exactly from / 25:29 7030 closed open to 7030 open close.
Laskin said: "we'll see most token demand going to open" (26:03).
26:03 So I think that um we'll see most token / 26:06 demand going to open.
Asked how much capital it takes to catch up to the frontier, Laskin said: "I think that now or let's say even six months ago is probably order billion so billions of dollars single digit billions" (09:18).
09:18 >> I think that now or let's say even six / 09:21 months ago is probably order billion so / 09:24 billions of dollars single digit / 09:25 billions.
On safety, Laskin said: "we have a few hundred safety researchers within closed labs that understand how these things work" (47:11), and "I have the belief that with enough eyeballs um most security and safety vulnerabilities become shallow as well" (47:01). The host had asked about the safety criticism of open models (44:44) and he was answering.
47:01 >> and I have the belief that with enough / 47:03 eyeballs um most security and safety / 47:06 vulnerabilities become shallow as well. / 47:09 >> So like the state of the world today is / 47:11 that we have a few hundred safety / 47:13 researchers within closed labs that / 47:15 understand how these things work
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep