Luis Buenaventura

Played around with @justinsuntron 's @BAI_AGI and genuinely surprised that people are still launching tool-less LLMs in 2026. 😬

Introducing Qwen3.6-Max-Preview, early preview of our next flagship model ·Improved agentic coding capability over Qwen3.6-Plus ·Stronger world knowledge and instruction following ·Improved real-world agent and knowledge reliability performance Smarter, sharper, still evolving.

Pitched an education idea here that I've been thinking about for awhile. Basic premise: the pricier the Claudes and ChatGPTs get, the further behind the rest of the world becomes. We need to plow more resources into cost-efficient inference ASAP!

another reason why i'll never get into Philippine politics: i hate doing cardio.

My favorite cloud host @Vultr just launched their own open-weights inference offering and the pay-as-you-go pricing is pretty bonkers. They're cheaper than @FireworksAI_HQ , @OpenRouter, @AskVenice, and (sans sales promos) Alibaba Cloud itself. (* Prices listed in screenshot are per-million tokens)

Banner rotation to "To The Moon" by @helloluis and part of the classic @MyCurioCards collection. You can never go wrong by acquiring art by legendary artists from classic collections.

If that's not a bear market signal I don't know what is.

Recently I've been hating on AI again. The proliferation of long-form slop essays on my feed has really turned me off of reading on FB, X, and LinkedIn. Whether it's a self-righteous political opinion exposing the truth that "no one is talking about", or a business dev reflecting on how the industry "hasn't just changed, it's being rebuilt", or a marketer talking about the "3 things people always get wrong" ... it's all become so predictable and tiresome. And the @ChatGPTapp illustrations! Oh my god. The lack of creative input is so apparent that they make me scroll faster just to avoid having to look at them. The whole point of including an illustration with your post is to attract more eyeballs to the writing, but sadly the more attention you give either of them, the less they deserve. Allow me to digress. We've known for awhile now that AI is a loss-leading industry, because the subscription pricing doesn't even come close to covering the inference costs. Basically, the more you use it, the more money they lose. It's one of those internal economic imbalances that is being addressed by pouring money into the machine ... you know, a bit like a Ponzi. So @OpenAI and @AnthropicAI aren't going to drop their rates anytime soon, if that's what you were hoping for. And they also can't increase the $20/month price tier because that's too strong of an anchor point. What will they do instead? Just give you LESS inference per dollar than they used to, and keep trying to up-sell you to the $100/mo level or more. Also, ads. But that goes without saying. @ChatGPTapp 5.5 Pro is now asking for $180 per million output tokens, @claudeai Opus 4.7 Fast is $165. And the new Anthropic tokenizer is supposedly going to burn thru 35% more tokens for the same output, so theoretically you get less for more. I mean, sure you can opt to use the slower/dumber versions, but success will accrue to the businesses with the most resources, won't it? (Not exactly the most novel insight but it bears repeating here.) At those price points, I would not be exaggerating when I say that a Filipino or Indian dev shop should seriously consider hiring back those junior developers they just fired. Which brings me back to why I'm hating on AI again. (And I confess that I've been going back and forth on this topic for the last 3 years.) The only thing that's currently cheap to inference is the slop on our feeds. All of the productive work -- the stuff that actually makes AI a step-change in academic research, engineering, and analysis -- is just gonna keep getting more expensive. I didn't really have a conclusion in mind when I started writing this essay (in markdown, I might add). I know my graphic sucks, but hey, at least you know it's me.

MiniPay is committing $1M in $CELO to back Mini App builders, paid out based on real transaction activity, not pitch decks 🛠️ ​​Most ecosystem grants fund proposals. MiniPay funds what's already working, supporting builders who know their markets best 🌍 Read more here: press.opera.com/2026/04/22/min…

📷 IN PHOTOS: Self-custodial stablecoin wallet MiniPay, built on Celo, an Ethereum level 2 blockchain network, allows developers to publish their apps in the Opera Mini browser. | via Michael Lance @Celo @minipay @RegenRene @celophdao @helloluis

Put together this fun little developer meetup for @Celo and @MiniPay tonight. The house is packed! ❤️🇵🇭

The @minipay and @Celo team has touched down in Manila! 🇵🇭 Looking forward to meet all the amazing founders and builders tomorrow!

Just a skill launching next week that helps you build a @minipay MiniPay in 30 minutes. And @Celo builders in HCMC were the first to test it. With exact feedback on how their idea could be a fit for MiniPay!

NFTs feature so many creative, brilliant people. Look at @helloluis. Around 2014, he was a Bitcoin startup founder in the Philippines. In 2017, he got the itch to be more creative, and started drawing doodles to celebrate Bitcoin price milestones. He kept posting on Twitter and grew an audience. Then his art started selling on T-shirts around San Francisco. He became known as Cryptopop. He went to San Francisco to raise capital. Then he met the Curio founders who asked him to create art for their cards. He has since sold at Christie's, authored three books, made art for CoinDesk, and continued to promote and drive crypto growth and innovation (especially in the Philippines). NFTs make it possible.

2030 is 3 wars, 2 financial crises, and 1 quantum computing scare away, so really who will even remember this statement

And @Celo quietly reaches 1.25B Total tx...🤫

Qwen 3.6 Plus from @Alibaba_Qwen is officially the first model on OpenRouter to break 1 Trillion tokens processed in a single day! At ~1,400,000,000,000 tokens, it’s the strongest full day performance of any new model dropped this year. Congrats to the Qwen team!

@helloluis 👉 Full Story: bitpinas.com/go/luis-returns

@jimmysong He got mad :-(

@jimmysong Your TG account just tried to phish me 😅 almost fell for it! We should definitely catch up at some point but IRL might be better!