# aya Published articles for aya. This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles. ## "Machine translation is still broken for most of the world's languages": Cohere builds non-reasoning for a reason DevFeed: ["Machine translation is still broken for most of the world's languages": Cohere builds non-reasoning for a reason]() Original publisher: [Read original article]() Author: Adrian Bridgwater Published: 2026-09-13T14:21:46Z Content type: news Language: en Sources: [The New Stack]() Topics: [Mixture of Experts (MoE)](), [cohere](), [Artificial Intelligence](), [Benchmark](), [Inference](), [qwen](), [gemma4](), [Google]() Tags: [ai](), [ai-infrastructure](), [ai-models](), [aya](), [benchmark](), [cohere](), [gemma](), [google](), [inference](), [large-language-models](), [mixture-of-experts-moe](), [model](), [open](), [qwen]() ### AI overview Cohere's North Small Translate is an open-weight mixture-of-experts machine translation model covering 50 languages. The article discusses its non-reasoning design, sovereign AI positioning, deployment options, efficiency claims, and reported WMT26 benchmark comparisons. ### Source excerpt Enterprise AI company Cohere announced North Small Translate last week, a mixture-of-experts (MOE) open-weight machine translation model that works across The post "Machine translation is still broken for most of the world's languages": Cohere builds non-reasoning for a reason appeared first on The New Stack. ## A Deepdive into Aya Vision: Advancing the Frontier of Multilingual Multimodality DevFeed: [A Deepdive into Aya Vision: Advancing the Frontier of Multilingual Multimodality]() Original publisher: [Read original article]() Author: Saurabh Dash; Yiyang Nan; Arash Ahmadian; John Dang Published: 2025-03-04T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [aya](), [vlm](), [cohere](), [Benchmark](), [datasets](), [Open Source](), [Large Language Model]() Tags: [aya](), [benchmark](), [cohere](), [datasets](), [language-models](), [llm](), [multimodal](), [open-source](), [performance](), [synthetic](), [technical](), [techniques](), [text-generation](), [vision](), [vlm]() ### AI overview Cohere For AI introduces Aya Vision, an open-weight multilingual and multimodal model family supporting language and vision understanding across 23 languages. The article describes its training techniques, benchmark results, open-weight releases, and image-processing architecture, including dynamic image tiling and latency improvements. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## A Deepdive into Aya Expanse: Advancing the Frontier of Multilinguality DevFeed: [A Deepdive into Aya Expanse: Advancing the Frontier of Multilinguality]() Original publisher: [Read original article]() Author: John Dang; Shivalika Singh; Daniel D'souza; Arash Ahmadian Published: 2024-10-24T00:00:00Z Content type: article Language: en Sources: [Hugging Face - Blog]() Topics: [aya](), [cohere](), [Large Language Model](), [synthetic-data](), [Training AI Models](), [Fine-tuning]() Tags: [ai](), [announcement](), [aya](), [cohere](), [comparison](), [llm](), [llms](), [open-source](), [research](), [synthetic-data](), [technical](), [training]() ### AI overview The Cohere For AI team presents Aya Expanse, an 8B and 32B multilingual model family designed to improve performance across languages. The article describes data arbitrage, multilingual preference training, safety tuning, model merging, evaluations across 23 languages, and the release of both models as open weights. ### Source excerpt We're on a journey to advance and democratize artificial intelligence through open source and open science. ## Blixt - A load-balancer written in Rust, using eBPF, born from Gateway API DevFeed: [Blixt - A load-balancer written in Rust, using eBPF, born from Gateway API]() Original publisher: [Read original article]() Author: The Kubernetes Authors Published: 2024-01-08T00:00:00Z Content type: article Language: en Sources: [Kubernetes Contributors Blog]() Topics: [eBPF](), [Rust](), [networking](), [Kubernetes](), [aya](), [Linux](), [Security]() Tags: [aya](), [ci](), [ebpf](), [kubernetes](), [kubernetes-clusters](), [linux](), [load-balancer](), [networking](), [rust](), [security]() ### AI overview The article introduces Blixt, a Layer 4 load balancer in Kubernetes SIG Network. Originally developed at Kong as an experiment for load-balancing Kubernetes ingress traffic with an eBPF dataplane, it was donated to Kubernetes and rewritten in Rust, with its control plane later moved to Golang. The project is intended to support CI and testing for Gateway API, especially TCPRoute and UDPRoute development. ### Source excerpt In SIG Network we now have a layer 4 ("L4") load balancer named Blixt . This project started as a fun experiment using emerging technologies and is intended to become a utility for CI and testing to help facilitate the continued development of Gateway API . Are you interested in developing networking tools in Rust and eBPF ? Or perhaps you're specifically interested in Gateway API? We'll tell you a bit about the project and how it might benefit you. History Blixt originated at Kong as an experiment to test load-balancing ingress traffic for Kubernetes clusters using eBPF for the dataplane. Around the time of Kubecon Detroit (2022) we (the Gateway API maintainers) realized it had significant potential to help us move our TCPRoute and UDPRoute support forward, which had been sort of "stuck in alpha" at the time due to a lack of conformance tests being developed for them. At the same time, various others in the SIG Network community developed an interest in the project due to the rapid growth of eBPFs use on Kubernetes. Given the potential for benefit to the Kubernetes ecosystem and the growing interest, Kong decided it would be helpful to donate the project to Kubernetes SIGs to benefit upstream Kubernetes. Over several months we rewrote the project in Rust (from C), due to a strong contingency of Rust knowledge (and interest) between us developing the project and an active interest in the burgeoning Aya project (a Rust framework for developing eBPF programs). We did eventually move the control plane (specifically) to Golang however, so that we could take advantage of the Kubebuilder and controller-runtime ecosystems. Additionally, we augmented our custom program loader (in eBPF, you generally write loaders that load your BPF byte code into the kernel) with bpfman : a project adjacent to us in the Rust + eBPF ecosystem, which helps solve several security and ergonomic problems with managing BPF programs on Linux systems. After the recently completed license review pro