Zyphra

Zyphra logo kit: download the mark, wordmark and square icons, and see how to use them.

Mixture-of-Experts (MoE) models improve efficiency by routing tokens to a subset of experts, but inter-GPU communication can consume up to 60.4% of training time.

Zyphra Research finds that local attention layers like sliding window attention (SWA) and gated linear attention create recency bias in hybrid models without explicit position encoding.

Zyphra Research shows that GPT-style decoder-only Transformers with 5M to 314M non-embedding parameters lose plasticity during multilingual continual learning.

Zyphra introduces PUFFER, a provenance-aware incremental fuzzy-deduplication system for continuously growing training corpora. PUFFER maintains a faithful, disk-resident MinHash locality-sensitive hashing history that can ingest new datasets, recover deterministically from interrupted jobs, and remo

Zyphra announced Zyphra Cloud, a full-stack AI platform on AMD powered by Tensorwave. The platform launches with Zyphra Inference, a serverless inference service for frontier open-weight models focused on long-horizon agentic workloads.

Zyphra today announced the availability of 15 megawatts of AMD Instinct MI355X GPU capacity through Zyphra Cloud, expanding the platform with software-driven infrastructure for bare-metal AMD deployments.

Joint collaboration between Zyphra, AMD, and IBM delivers ZAYA1, the first large-scale Mixture-of-Experts foundation model trained entirely on an AMD platform using AMD Instinct MI300X GPUs, AMD Pollara networking & ROCm software.

ZAYA1-8B delivers reasoning, mathematics, and coding performance competitive with models many times larger, achieving high intelligence density with under one billion active parameters trained on full-stack AMD infrastructure.

Open roles at Zyphra, a product-forward research lab in San Francisco building the full stack for open intelligence.

The role of Data Engineer - Multimodal Systems involves developing and improving Zyphra’s datasets and data pipelines.

The Machine Learning Systems Administrator / DevOps Engineer maintains infrastructure for Zyphra’s model training and operations.

The Research Engineer - Agency and Reasoning contributes to Zyphra’s Agency and Reasoning Team on research projects.

Full stack open superintelligence. Zyphra offers multisilicon infrastructure, optimized and customizable open frontier models, and a productivity agent so companies can choose, customize, and control their intelligence.

How Zyphra collects, uses, and discloses personal information in connection with the MAIA beta program, including Google user data and retention.

The terms that govern participation in the MAIA beta program, including evaluation use, feedback, confidentiality, and disclaimers.