Very much enjoyed working with and learning from a distinguished group of physicists turned ML scientists. Happy to announce my very first ML paper. Let the journey begin… https://t.co/gw47r3QMKl

In collaboration with @AMD and @IBM, we @ZyphraAI are sharing ZAYA1-base! The first large-scale model on an integrated AMD hardware, software, and networking stack. ZAYA1 uses Zyphra’s novel MoE architecture with 760M active and 8.3B total params. Tech paper and more below👇

Introducing the compact, dense versions of Qwen3-VL — now available in 4B and 8B pairs, each with both Instruct and Thinking variants. ✅ Lower VRAM usage ✅ Full Qwen3-VL capabilities retained ✅ Strong performance across the board Despite their size, they outperform models

Check out the new paper from @ZyphraAI

A year ago Ross Girschik (the object detection GOAT) gave a talk on "real tasks" (the end goal of an ML system) vs "fake tasks" (the intermediate tasks we created to achieve the real tasks). Most vision tasks, like classification and detection, are fake. (1/5)

Today, we at @OpenAI achieved a milestone that many considered years away: gold medal-level performance on the 2025 IMO with a general reasoning LLM—under the same time limits as humans, without tools. As remarkable as that sounds, it’s even more significant than the headline 🧵

Products with extensive/rich UIs lots of sliders, switches, menus, with no scripting support, and built on opaque, custom, binary formats are ngmi in the era of heavy human+AI collaboration. If an LLM can't read the underlying representations and manipulate them and all of the

Zyphra is releasing our first reasoning model, ZR1-1.5B. This small but powerful reasoning model excels at both math and code, making it one of the best models in these categories for its size. It also uses 60% less reasoning tokens than comparable models. 🆓Apache 2.0 license.

~ 10K people are working in US quantum tech, this is ~ 5 years worth of US physics PhD production (not all are physics PhDs), estimated total expense/yr ~ $5 billion, income ~ zero How long can this continue? If/when quantum winter comes what would happen to these QC workers?

Today, we're excited to announce a beta release of Zonos, a highly expressive TTS model with high fidelity voice cloning. We release both transformer and SSM-hybrid models under an Apache 2.0 license. Zonos performs well vs leading TTS providers in quality and expressiveness.

We @Zyphra previously described our preliminary RAG system that achieved SOTA performance on the HashHop long-context task. We are now excited to share our paper presenting a newer version of this RAG system that achieves SOTA results across multiple long-context benchmarks.

Oh this is great! Who said quantum computers had to work with qubits? Turns out it's possible to design a quantum algorithm for factoring that only requires 3 quantum oscillators and a single qubit! (1/4)

We’ve been hard at work with @AMD to optimize training for AMD GPUs. Today, we’re sharing a critical milestone towards this goal: FlashAttention-2 (FA2) and Mamba-2 backward kernels on AMD MI300X that surpass NVIDIA H100. We @ZyphraAI are the first to achieve this.

Very informative post from Scott Aaronson about Google's Willow quantum announcement yesterday:

Damned impressive, if taken at their word. "Logical computation demonstrated with a neutral atom quantum processor” [claiming up to 28 log. qubits in 256 phys., implementing real error-corrected algorithms] https://t.co/SGkg8YVQs1 #Quantum #QuantumComputing #AtomComputing

Remember exercise pages from textbooks? Large-scale collection of these across all realms of knowledge now moves billions of dollars. Textbooks written primarily for LLMs, compressed to weights, emergent solutions served to humans, or (over time) directly enacted for automation.

Did you know that a leading open LLM dataset, DCLM, is ~80% duplicates? We discovered this while making Zyda2. Although performance seems fine on evals, downstream effects are less clear. So here are the 750B tokens of deduped, quality tokens from DCLM: https://t.co/WZIDfLoo7l

The Nobel Committee recognizes profound contributions from Physics to ML / AI. There's a lot more where that came from. We are in an era where an increasing number of physicists are making important contributions to ML / AI, and even more are needed going forward.

Cisco Quantum Summit is happening right now! Tune in if you are interested: https://t.co/2v9zD0QTw8

A major step in error correction @GoogleQuantumAI : Pushing the surface code to the next level Below threshold: Distance-7 logical qubit, 0.0014 error per cycle, >2x better than physical qubits. https://t.co/N0UKUFtjWn

What do graph theory, many-body physics, the golden ratio, and Fibonacci anyons have in common? In our experiment, arXiv link below, I’m excited how a very fundamental graph problem – https://t.co/YzzlTXMrce 1/...