Zyphra Technologies Inc.
| File:Zyphra logo.png | |
| Private | |
| ISIN | 🆔 |
| Industry | |
| Founded 📆 | |
| Founders 👔 | Krithik Puthalath (Chairman & CEO) Beren Millidge (President) Tomas Figliolia (Chief Architect) Daniel Martinelli |
| Headquarters 🏙️ | |
Area served 🗺️ | |
Key people |
|
| Products 📟 |
|
| Members | |
Number of employees | 65+ (2025) |
| 🌐 Website | www |
| 📇 Address | |
| 📞 telephone | |
Zyphra is an American, open source artificial intelligence company based in San Francisco, California.
The company operates as a full-stack AI research and product lab that develops foundation models, infrastructure, and agentic AI applications.[1]
Zyphra was founded by Krithik Puthalath, Beren Millidge, Tomas Figliolia, and Danny Martinelli.
Zyphra is building foundation models based on a scalable general architecture designed for long-term memory, multimodal world models, and recursive self-improvement with continual learning.[2]
In addition to its foundation models, Zyphra has two main products: Zyphra Inference Cloud (an inference platform for open-source models) & Maia (a multiplayer general-agent assistant). [3]
In 2025, Zyphra raised a Series A funding round at a $1 billion post-money valuation, led by Jaan Tallinn, who previously led Series A rounds for DeepMind and Anthropic.[4]
History
Zyphra was founded in 2021 by Krithik Puthalath with the aim of pursuing human-aligned superintelligence.[5]
Prior to founding Zyphra, Puthalath worked at Cambridge Quantum Computing (Quantinuum), where he worked on algorithms, compilers, and applications in quantum machine learning.[6] This experience informed Zyphra’s thesis on integrating next-generation learning algorithms, neural network architectures, and hardware-aware systems design.
In its early formation, Puthalath recruited a small group of researchers and operators to serve as co-founders of the lab. The founding team included Beren Millidge, a researcher with postdoctoral training at the University of Oxford. Prior to Zyphra, Millidge was the co-founder of AI safety lab Apollo Research[7] and Chief Scientist at Conjecture, an AI alignment lab backed by Nat Friedman, Daniel Gross (businessman), Patrick Collison, Andrej Karpathy and Sam Bankman-Fried.[8] Other founding members of Zyphra included Tomas Figliolia and Danny Martinelli.
During its initial pre-seed phase, Zyphra focused on exploratory and fundamental research. As part of this work, the company collaborated on research into learning algorithms, including a peer-reviewed publication[9] in Nature (journal) on probabilistic algorithms and systems co-authored with John M. Martinis, a Nobel Prize laureate in physics.[10]
In early 2024, Zyphra published research on layer-pruning strategies for pretrained large language models in collaboration with Meta, Sequoia Capital, and Cisco, examining efficiency and performance trade-offs in large-scale models.[11] During the same period, Zyphra announced BlackMamba, an open-source hybrid architecture combining structured state-space models with mixture-of-experts techniques.[12]
In April 2024, the company introduced Zamba, a 7-billion-parameter hybrid (SSM-attention) foundation model.[13] Throughout mid-2024, Zyphra published additional work on retrieval-augmented generation conversational agents, long-context memory systems, and efficient model deployment.
In October 2024, Zyphra released Zamba2-7B, outperforming Mistral AI, Gemma (language model) and Llama (language model) models in both quality and performance.[14] During this period, Zyphra also introduced Tree Attention, an algorithm designed to improve the efficiency of large-scale inference.[15]
Following additional financing in late 2024, Zyphra published research on frontier training kernels for AMD Instinct MI300X accelerators, mixture-of-experts systems, and hardware-optimized training and inference pipelines.[16]
In February 2025, Zyphra released Zonos-v0.1, an open-weight text-to-speech model.[17]
In April 2025, Zyphra introduced ZR1, its first reasoning-focused language model, which was post-trained on DeepSeek’s R1 model to improve multi-step reasoning and inference behavior.[18]
After completing its Series A funding round in June 2025, Zyphra entered a period of expanded infrastructure partnerships and frontier-scale model training.
In October 2025, IBM and AMD announced a multi-year collaboration with Zyphra to provide advanced AI infrastructure on IBM Cloud, including large clusters of AMD Instinct MI300X accelerators and integrated AMD networking hardware intended to support the training of large-scale and multimodal foundation models. Under this agreement, Zyphra was positioned to utilize one of the largest dedicated GPU clusters deployed by IBM Cloud and AMD for large-scale training workloads.[19]
In November 2025, Zyphra trained ZAYA1, a large-scale mixture-of-experts (MoE) foundation model trained entirely on AMD platform using AMD Instinct MI300X GPUs and AMD Pensando networking technology.[20]
Funding
Following its founding, Zyphra operated as a bootstrapped lab with pre-seed funding by Krithik Puthalath.[21]
In 2023, Zyphra raised an $11 million seed funding round with participation from Intel Capital, Future Ventures, and Bison (spinout from the Bill Gates family office).[22] Future Ventures was co-founded by Steve Jurvetson, who has served on the boards of SpaceX and Tesla, and has been involved with other companies founded by Elon Musk, including Neuralink and xAI.[23]
In June 2025, Zyphra completed a $100 million Series A funding round at a $1 billion post-money valuation. The round was led by Jaan Tallinn, an early investor in Artificial General Intelligence research companies who previously led Series A funding rounds for DeepMind and Anthropic. Follow-on investors included AMD, Gaia (spinout from SAP), and AlphaJWC Ventures, a Southeast Asia–based venture capital firm.[24]
See also
References
- ↑ "About Zyphra". Zyphra.
- ↑ AI, the Brain, and Our Future. Youtube.
- ↑ "Zyphra Taps IBM, AMD To Build Next-Gen AI Superagent". Yahoo Finance.
- ↑ "Open-Source AI Startup Zyphra Targets $1 Billion Valuation". TheInformation.
- ↑ "Zyphra releases Zamba, an SSM-hybrid foundation model to bring AI to more devices". VentureBeat.
- ↑ Tom Benjamin & Krithik Puthalath: Cambridge Quantum Computing - Focus Areas. Youtube.
- ↑ "Beren Millidge - About me". Beren Millidge.
- ↑ "We Are Conjecture, A New Alignment Research Startup". Conjecture.dev.
- ↑ Aadit, Navid Anjum; Grimaldi, Andrea; Carpentieri, Mario; Theogarajan, Luke; Martinis, John M.; Finocchio, Giovanni; Camsari, Kerem Y. (2 June 2022). "Massively parallel probabilistic computing with sparse Ising machines". Nature Electronics. Springer Nature. 5: 460–468. doi:10.1038/s41928-022-00774-2.
- ↑ "2025 Nobel Prize Resources – Meet the Laureates". AIP.ORG. October 7, 2025. Retrieved October 7, 2025.
- ↑ "The Unreasonable Ineffectiveness of the Deeper Layers". Zyphra.
- ↑ "Zyphra Open-Sources BlackMamba: A Novel Architecture that Combines the Mamba SSM with MoE to Obtain the Benefits of Both". MarkTechPost.
- ↑ "Training Zamba: A Hybrid Model Master Class with Zyphra's Quentin Anthony". Cognitiverevolution.ai.
- ↑ "Zyphra Launches Zamba2, An Efficient And Faster Small Language Model". Forbes.
- ↑ "New Breakthrough in GPU Optimization! 'Tree Attention' Accelerates Inference of 5 Million Long Texts by 8 Times". Latest AI News.
- ↑ "Training Transformers and Hybrid models on AMD Instinct MI300X Accelerators". AMD ROCm.
- ↑ "This open text-to-speech model needs just seconds of audio to clone your voice". TheRegister.
- ↑ "Zyphra's tiny ZR1-1.5B AI model beats Claude 3 and Qwen72B in math, code benchmarks". News9live.
- ↑ "IBM and AMD Collaborate with Zyphra on Next Generation AI Infrastructure". IBM.
- ↑ "AMD Powers Frontier AI Training for Zyphra". AMD.
- ↑ "Zyphra Financials". Crunchbase.
- ↑ "What To Make of OpenAI's New Board; The Startup Tackling Karpathy's Vision". TheInformation.
- ↑ "Investors in Talks to Help Elon Musk's xAI Raise $3 Billion, Adding to Industry Arms Race". Wall Street Journal.
- ↑ "Zyphra Overview". PitchBook.
This article "Zyphra" is from Wikipedia. The list of its authors can be seen in its historical and/or the page Edithistory:Zyphra. Articles copied from Draft Namespace on Wikipedia could be seen on the Draft Namespace of Wikipedia and not main one.
