/AI3h ago

Coinbase's Brian Armstrong predicts 80% of AI workloads will move to 99% cheaper specialized models within 18 months

Only high-level orchestrators will require expensive frontier models.

2632.2K1681.2K479.6K
#488kache
Yash Patil
Original post
Yash Patil@ypatil125#1235inAI

This is exactly right.

People are starting to look for cheaper model alternatives and realizing two things at once: open-source models are already very good, and the ability to train and serve them efficiently at scale can change the economics pretty meaningfully.

Tokens are still being subsidized, demand is ramping quickly, and the compute crunch is likely to persist. That will push companies toward using the right model for each task instead of defaulting to the most expensive one.

We’re still early, but I expect open-weight adoption to accelerate much faster than most people think.

7:24 PM · Jun 7, 2026 · 4K Views
Sentiment

Positive users agree with Brian Armstrong's take that open-source inference providers are slashing AI API costs, while negative users insult him personally and raise unrelated grievances like past crypto projects.

Pos
29.6%
Neg
70.4%
23 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS5.4KBOOKMARKS7LIKES64REPLIES12
kache@yacineMTB

The frontier models are pretty retarded, and are only useful for chainsaw level tasks that don't require that much intelligence like mathematics

Brian Armstrong@brian_armstrong

Good take

My guess is - demand for intelligence is near infinite - but 80% of workloads will be running on 99% cheaper models within 12-18 months - 20% of workloads will still run on latest gen models where IQ maxing is important (scientific breakthroughs, higher level ochestrator agents?) - rough analogy might be what % of macbooks or gaming PCs sold have the maxed out specs for CPU/GPU, prices are falling much faster than Moore's law here though - this leads me to think the limiting factor will be energy and compute, not better models

At Coinbase we're working hard on routing prompts to cheaper models where appropriate, and in some cases have been able to keep costs roughly flat, while token usage continues to grow exponentially.

1hViews 5.4KLikes 64Bookmarks 7
RETWEETS4
ST0RM@0xStorm

@brian_armstrong Seeing this live through convos with builders in our private beta - we built a reasoning engine that helps cheaper models match or beat frontier model performance

Cost savings have been ~90% for teams switching over

If this is a focus would love to connect

2hViews 863Likes 13Bookmarks 2
kache@yacineMTB

Idk. The more this plays out the more I realize that the quality of intelligence isn't really what it seems to be. To me, AGI is the evolutionary process, the spread, the sampling, the search, the spamming

kache@yacineMTB

The frontier models are pretty retarded, and are only useful for chainsaw level tasks that don't require that much intelligence like mathematics

1hViews 1.6KLikes 29Bookmarks 3
Surplus Intelligence@AskSurplus

@brian_armstrong You can run workloads ~80% cheaper on Surplus right now

https://www.surplusintelligence.ai/

2hViews 771Likes 19Bookmarks 1
kache@yacineMTB

It's just not that impressive you know? Not as impressive as some humans that I know. Some very, very smart humans, remarkable people I've had the opportunity to meet

kache@yacineMTB

Idk. The more this plays out the more I realize that the quality of intelligence isn't really what it seems to be. To me, AGI is the evolutionary process, the spread, the sampling, the search, the spamming

1hViews 1.6KLikes 23Bookmarks 0
Crypt0_AI@Crypt0_AI

@brian_armstrong Hey Brian - @openservai has already solved this problem through their reasoning layer. One line of code that makes current LLMs more efficient, cheaper, and reliable. Please take a few minutes to read this article! They're on your Basechain afterall!

2hViews 711Likes 11Bookmarks 2
kache@yacineMTB

One of the things that I cope about: it takes barely any effort to have a creative spark, and the cost of building something that has that would be too much to be super human. And we like it, so why? So automate the cheap drudgery, so we can create!

kache@yacineMTB

It's just not that impressive you know? Not as impressive as some humans that I know. Some very, very smart humans, remarkable people I've had the opportunity to meet

1hViews 882Likes 15Bookmarks 1
Tommy@Shaughnessy119

@brian_armstrong Brian thank you for reading!

I would love to discuss AI and Crypto with you sometime. Happy to do a call, or I will come to you in person!

We have a few exciting things we are working on that I would love to run by you

3hViews 1.2KLikes 13Bookmarks 1
Jeffrey Stewart@UrgentSpeed

@brian_armstrong Running models will get 99.9% cheeper...AND...demand will grow 1000X. The question is what happens sooner.

2hViews 1.3KLikes 5Bookmarks 1
Tomas Hernando Kofman@tomas_hk

@brian_armstrong Thanks for the s/o @yenkel . @brian_armstrong, have been working on routing for 2 years, working with a bunch of F100s on routing for coding agents. We should talk! http://notdiamond.ai

2hViews 489Likes 6Bookmarks 1
白小白TIM@ClarisseA3055

@brian_armstrong Agreed. On-chain data shows GPUcompute tokens like $RENDER and $GPU have seen 340% active wallet growth since AI mania. Compute scarcity = next big crypto infra play.

3hViews 1Likes 5
Anil Murty ⟁@anilmurty_ai

@brian_armstrong TokenMaxxing needs to evolve to Token Efficiency Optimization. Working on an open source product that achieves that with 4 well researched techniques: http://TokenJam.dev

2hViews 304Likes 1Bookmarks 1
yenkel@yenkel

@brian_armstrong > At Coinbase we're working hard on routing prompts to cheaper models where appropriate

you should chat with @tomas_hk

3hViews 459Likes 5
staysaasy@staysaasy

@brian_armstrong What are you using for your cheaper model stack

3hViews 468Likes 6
jacob@jsnnsa

i think this consolidation happens as a capability generation ages.

IMO the last capability generation started with sonnet 3.5 and is likely to end soon. will you actually want coinbase eng prompts going to GLM 5.1 if next gen is significantly better? (like GPT 4 to sonnet 3.5 better, not opus 4 to 4.1)

I suspect we’ll again be back to “OSS super far behind, not using frontier models is a competitive disadvantage” rather soon

2hViews 467Likes 2
Aaron Decker@ardninja

@brian_armstrong A lot of people said this for the last 2 years every year and it hasn't been true so far.

2hViews 447Likes 2
narendra 🎈 🏃🏻‍♂️@narendra

@brian_armstrong @grok what is your best guess at the % likelihood that @brian_armstrong's guess is accurate.

26mViews 43
Soph 🧲@aussiesoph

@brian_armstrong Until you sort out $WLUNA anything you have to say means fuck all

1hViews 35Bookmarks 1
BH@billyhu6

@DamirWallener @brian_armstrong This is misunderstanding evolution and its "goals" vs human desires. Evolution is blind, directed by the environment towards fitness, reproduction. Life without any intelligence is abundant. Humans can contravene evolution (e.g. contraception) and have separate desires, demand...

1hViews 8Likes 1
InvestingVault@InvestingVault

@brian_armstrong I wish you just improved @CoinbaseSupport. By far the worst in the industry

49mViews 11Likes 1
Load more posts