
terminal romantic @cloudflare
nyc
Joined January 2019
- Tweets3.4K
- Following1.7K
- Followers10.8K
- Likes24.1K
me, my best friend, my ex-boyfriend, my ex-boyfriend’s ex-girlfriend, my ex-boyfriend’s ex-best friend, my ex-boyfriend’s ex-best friend’s ex-girlfriend (whomst my ex-boyfriend is now dating). we’re all in tokyo. itadakimasu~
33
1,117
Ohhh have you tried Clairic? Actually better for this use case developers.cloudflare.com/wo…
mourning who i was mourning who you are mourning who i will be mourning who i can’t. the most nostalgic girl on the planet grieves a future she hasn’t lived yet
16
1,368
inkling !!
We swapped over from GLM 5.3 Flash to @thinkymachines Inkling 2 Small in the app I’ve been working on WinEvents.io (Windows event observability with an agent) and it’s been excellent and super fast. Thanks again to @dok2001, @zatlyn and @michellechen for your help and the @Cloudflare Basin team.
i know my vlog made everything seem rosy but we did 11 customer meetings in 3 cities across 4 days in china. all while working our day (night) jobs. gotta make it look easy
michelle retweeted
Cloudflare just dropped Clairic on @huggingface ☁️✏️🧾
Clairic goes beyond decision models: it actually extracts information on a single forward pass, unlocking ultra fast system one decision-making process
▶️ huggingface.co/spaces/huggin…
kevin just forced me to translate kubernetes terms in chinese and the best i could do was say helm chart really slowly
michelle retweeted
✴️ BENCHMARKED: CLOUDFLARE WITHOUT THE CLOUD
@Cloudflare's new open-weight document reader, Clairic-flash 9B, running fully local on my 64GB M5 Pro Mac mini.
As promised, here's tonight's benchmark.
- It correctly extracted all 34/34 pieces of information from an image of a receipt, a pdf of an invoice and a form
- I ported it to MLX using @SpaceXSI and it's blazing fast on Apple silicon
- 40.5 tok/s at 4-bit (3.2x faster than with PyTorch!)
- 36 tok/s even with a long context size (16K)
- Model loading time: 3.2 s
- Memory usage: 7.2 GB
- 33/34 fields from the same documents' text on MLX
A local document reader LLM that actually works.
Huge release from Cloudflare 🧡
JUST IN: @Cloudflare is a neolab now and it's shipping open weights under Apache 2.0
Clairic-flash is the 9B, fine-tuned from Qwen3.5-9B. The BF16 weights take up about 20GB and I was able to run inference on my 64GB Mac with room to spare.
Every extracted field also comes with a confidence score, which will come in handy (esp. for receipts and invoices).
Will post benchmark results tonight.
clairic is redesigned and now open-weights! we went back to the drawing board and overhauled the architecture for it to excel at extraction.
new:
- clairic (27B) and clairic-flash (9B)
- image extraction greatly improved
- PDF support
- weights release!
blog.cloudflare.com/open-cla…
thank you for all the feedback and experimentation with clairic while it was in preview. hope you enjoy the improved capabilities, and can’t wait to see what new things get cooked with the weights release
huggingface.co/Cloudflare/cl…
spite-driven development. i love proving myself and others wrong
People ask me what keeps me at Cloudflare. Its pretty simple my desire to prove @eastdakota wrong which he constantly use against me and be surrounded by a team that has the same mission.
Find a manager that knows your “why”. This applies to teachers, coaches, or anyone that is trying to make you better.
michelle retweeted
And now, a year later, every major harness has implemented it... and they all kept the name.
It's great to see everyone adopting Code Mode! Like most good things @Cloudflare, it started with a blog post by @KentonVarda. He wrote it before the code, and we never got around to figuring out a better name.
i think the theme for today is that i love the industry i love that other companies are launching extraction models (@liquidai) or supporting extraction endpoints (@OpenRouter!) and i love that we get the privilege of sitting in the center of it all. everyone wins when agents get more reliable and autonomous!!
Classifiers need categories up front, but you can't pre-write categories for an email you've never read.
Michelle Chen (@michellechen) on Clairic, @Cloudflare's generalized extraction model: it returns a confidence score and a character span.
thursdai.news/nov-12
i kept getting questions on how we released clef so quickly after jev, but i strongly believe that we’re mostly reaping the benefits of asymmetric effort: new ideas are difficult to invent and easy to copy.
@typesafeai has done an incredible job in pioneering this concept of system one models. i also strongly believe that it is not a zero-sum game! it’s almost a duty for all of us to solve difficult hardware constraints through software, and system one models allow us to offload some of the work we’d regularly send to LLMs towards smaller, faster, and less compute-intensive models.
we’re also still challenging ourselves to invent in this space as well, with the launch of clairic and more. i’m excited for the future of cloudflare as a neolab and hope you are too. thank you to @ricmac for sitting down with me on this story!
I've been following @Cloudflare since it began 17 yrs ago. Its latest move into creating System One models is fascinating, and when I sat down with @michellechen at @aiDotEngineer Code, I discovered there are more models to come — incl reasoning models! latent.space/p/cloudflare/





















