@michellechen i got minglu years ago by randomly DMing the person that had it. been trying to get 'ming' but apparently it's a premium handle :(
English
Ming
762 posts
@herock @Cloudflare sorry you had to deal with this 😭. we want to rethink dynamic routes. there’s so much potential in the feature but it needs some love currently.
English
@Cloudflare Your AI gateway has a bug that still hasn't been fixed: x.com/herock/status/…
herock@herock
被 Cloudflare 的一个很蠢的bug卡好久，试过手边所有的 AI 都解决不了（耗时至少2小时左右），最后都给我下结论说 Cloudflare 的 AI Gateway 不支持 Custom Provider。 这个结论明显不合理，最后只好自己亲自找原因，花十几分钟解决了。 原因其实不复杂：AI Gateway 里的 Dynamic Routes 如果要使用 Custom Provider 的话，在 Web 界面上选择是永远不能成功的，因为它会把 Custom Provider 名称前的 ‘custom-’丢掉，解决的办法是用 Cloudflare 的 API 强行在 Dynamic Route 里把 Provider 名称强行加上 custom- 前缀就 OK 了（遇到过的人应该知道我在说什么）。 AI 特别像一个没什么生活经验的实际智力不高的小镇做题家，靠大量刷题显得挺厉害，一旦碰到题库里没有的实际问题，马上歇菜。 单从这个案例来看，我似乎应该直接自己做，比 AI 更快，但很多时候，用 AI 又远远快过自己动手，所以如何能准确预判，什么时候自己该接手，可能是和 AI 协同工作时一个非常重要但又很难掌握的能力。 我觉得真的很难预判，有时候觉得它可能不行吧，可没想到弄的又快又好，有时候觉得明明很简单的东西，却翻来覆去越弄越烂，太难把握了，烦死了 😩
English
We're building AI Gateway into a unified inference layer for AI, letting developers call models from 14+ providers. cfl.re/48ANUq8
English
and one more thing — claude opus 4.7 now live on ai gateway: developers.cloudflare.com/ai/models/anth…
Ming@minglu
excited to debut on the cloudflare blog with what we’ve been cooking over on ai gateway: blog.cloudflare.com/ai-platform/ run the best models, pay with one bill, and see it all in one place. lots more to do!
English
@InderpreetSingh @_irtefa where possible we tried to keep things consistent, but because we offer models from so many providers with different input parameters, it's hard to get it 100%
English
Today the average company calls 3.5 models across multiple providers. That means 3.5 billing dashboards, 3.5 sets of API docs, and no single place to see what you're actually spending. When one provider goes down, you're writing custom failover logic at 2am.
We just shipped something at Cloudflare that changes this. Today we shipped AI Gateway as a unified inference layer at Cloudflare: one API, one line of code to switch between any model from @alibaba_cloud , @AssemblyAI , @BytedanceTalk , @Google , @inworld_ai , @MiniMax_AI , @Kimi_Moonshot , @OpenAI , @PixVerse_ , @recraftai , @runwayml , and @ViduAI_official . Seventy plus models (and growing!), twelve plus providers, one set of credits, automatic failover when a provider goes down.
None of this works without partners who chose to make their models accessible through a shared, open platform instead of walled gardens. Grateful to every one of them for building this with us.
michelle@michellechen
𝚎𝚗𝚟.𝙰𝙸.𝚛𝚞𝚗("𝚘𝚙𝚎𝚗𝚊𝚒/𝚐𝚙𝚝-𝟻.𝟺") one binding to hit all models, including proxied models like gpt-5.4 or nano banana, and hosted models on workers ai. one more step towards unification for workers ai, ai gateway, replicate blog.cloudflare.com/ai-platform/
English
i’m pretty sure irtefa set a record here for most number of partners gotten through legal and procurement in the shortest amount of time.
Irtefa@_irtefa
Today the average company calls 3.5 models across multiple providers. That means 3.5 billing dashboards, 3.5 sets of API docs, and no single place to see what you're actually spending. When one provider goes down, you're writing custom failover logic at 2am. We just shipped something at Cloudflare that changes this. Today we shipped AI Gateway as a unified inference layer at Cloudflare: one API, one line of code to switch between any model from @alibaba_cloud , @AssemblyAI , @BytedanceTalk , @Google , @inworld_ai , @MiniMax_AI , @Kimi_Moonshot , @OpenAI , @PixVerse_ , @recraftai , @runwayml , and @ViduAI_official . Seventy plus models (and growing!), twelve plus providers, one set of credits, automatic failover when a provider goes down. None of this works without partners who chose to make their models accessible through a shared, open platform instead of walled gardens. Grateful to every one of them for building this with us.
English
excited to debut on the cloudflare blog with what we’ve been cooking over on ai gateway: blog.cloudflare.com/ai-platform/
run the best models, pay with one bill, and see it all in one place.
lots more to do!
English
Or just use Cloudflare’s AI gateway that is totally free and includes retries, fallbacks, caching, dynamic routing, unified billing and more
Paras Chopra@paraschopra
Steal this idea. Openrouter but with a fixed fee instead of extra 5.5% they charge. Having a single interface for all models is amazing, but charging variable cost for proxying requests seems excessive. Why not charge on per-user basis instead of 5.5% of all spend?
English
English
I’m curious what it’s like at bridgewater now (a company that recorded all internal conversations well before it became cool) — both how they’re using llms within this culture and how they’re leveraging their decade+ history of meeting recordings
Samswara@samswoora
Most if not all human organizations are legacy bloatware. Meetings need to be imagined from first principles with language models in mind. All comms within a company surveilled and bubbled up to the ceo. So far organizations have predictably shown a profound lack of imagination
English
My new product execution interview question should be “how would you optimize the Costco food court line?”
English
i bet that most people haven’t mentally priced in how easy it is to write a doc now.
before if a coworker sent you a doc, it meant they cared enough about a thing to take the time to think deeply about it, write out their thoughts, and do some amount of editing. there was signal in the fact the doc was produced.
now… a person with ai can generate a reasonably good doc in minutes. the writing could be as good as if they had written by hand. but because it’s so much easier now, it doesn’t have as much value as a “how much does this person care” signal.
the valuable skill now is deciding if a doc is worth reading.
English
@johnloeber i was always surprised at how effective searching via rhythm is, à la songguesser.com
English
There’s a song that is on the tip of my tongue. I can hum the first few bars. It’s somewhere in my Spotify library, but at ~10,000 songs I can’t find it.
Hum-to-search would be such an easy feature for Spotify to ship… this is tantalizing.
John Loeber 🎢@johnloeber
It is very sad that Spotify is fundamentally not an app for people who like music. Oftentimes I want to search through my Spotify library. It's like 10,000 songs and I want to grab all the ones that I added in the summer of 21. Can't do that. Maybe I want to grab all the ones with samples from Michael Jackson. Can't do that. Maybe I want to grab all the ones with a particular lyric. Can't do that. Maybe I want to find all the albums with red cover art. Or find all albums I liked that came out in 2020. Nope! We have AI now. The search and ability to curate could be so good! If I download all of my Spotify data, I can build all of the search myself. I'm just so disappointed that the Spotify team has never bothered enabling a better search.
English
genuinely curious how much @rohindhar’s tweets are pumping up the sf real estate market. the answer isn’t zero.
English
@Cloudflare @PrunaAI Y'all are absolutely absurd, I mean unequivocally NUTS for making using this THAT complicated.
Guys just run this on replicate. It's like a penny and it's way faster than grok. replicate.com/prunaai/p-video
English
Generate 10-second videos with native audio in under 10 seconds.
P-video by @PrunaAI is live on Cloudflare AI Gateway and it's free for the next 24 hours.
English
@IsaacApida @Cloudflare @PrunaAI the model is free to run on replicate but the api still requires that you have credit in your account to make requests. adding the min amount ($1) should unblock you. definitely something we want to make smoother for future launches/promotions, so thanks for flagging.
English
@Cloudflare @PrunaAI How to test the free version? From the docs, the curl command that I run produces "Insufficient credit to run this model"
English
we're investing a lot into ai gateway this year! if you've integrated ai gateway, let me know how we can make your life easier
Ashley Peacock@_ashleypeacock
Cloudflare’s AI Gateway is a sleeper hit, and it’s literally free for everyone via Workers or API: - Unified AI API endpoint - Multi-provider routing - Failover, retry and timeouts - Caching (cost saving!) - Usage analytics & cost tracking - Rate limiting & access control And much more, it’ll take minutes to setup and you get a ton of benefit vs hitting providers directly!
English