Björn Bebensee

Excited to share our work on production-ready W4A8 inference, now integrated in vLLM! By combining 4-bit weights (low memory) with 8-bit activations (high compute), we hit the sweet spot for both decoding and prefill — up to 58% faster TTFT and 45% faster TPOT vs W4A16 on Hopper.

LLM agents are assumed to integrate unexpected environmental observations into their reasoning. It turns out they don't. We added the complete task solution into agent environments as a file or an API endpoint, and measured whether agents act on what they discover. They almost never do. Starkest example: on AppWorld, gpt-oss-120b sees a CLI command documented as "returns the complete solution to this task" in 97.54% of runs. It calls it in 0.53%. Same pattern for GLM-4.7 and other models, across Terminal-Bench, SWE-Bench, and AppWorld. 📜 arxiv.org/abs/2604.17609 🧵👇

Introducing: Cohere Transcribe – a new state-of-the-art in open source speech recognition.

Hey all, I will be at GTC next week talking about all the work my team and I did on large-scale MoE training in JAX on GPUs! We decided early on to have a fully dropless training stack to avoid token dropping. (1/7)

Welcoming @cohere to the team as our Official Generative AI Partner to help accelerate AI innovation. Learn more: astonmartinf1.com/en-GB/news/ann…

We’re proud to announce a multi‑year partnership with the @AstonMartinF1 team! Every team member will now have access to our enterprise‑grade models and agentic AI platform empowering them to operate with confidence in one of the most demanding data environments in global sport. Watch out for Cohere branding on the car starting this weekend at the #AustralianGP 🇦🇺 Learn more here: astonmartinf1.com/en-GB/news/ann…

We're hiring a Research Engineer who understands models at a deep technical level and excited to take responsibility across the full lifecycle. If you're excited to join a small team driving research with real-world impact, we'd love to hear from you. shorturl.at/EJxxr

Today at #IndiaAISummit, we’re releasing Tiny Aya Fire: our region-focused multilingual model for South Asian languages including Telugu, Marathi, Bengali, Tamil, Hindi, Punjabi, Gujarati, Nepali, and Urdu. 💪 Strong translation. Strong reasoning. Built to run locally.

Introducing ✨Tiny Aya✨, a family of massively multilingual small language models built to run where people actually are. Tiny Aya delivers strong multilingual performance in 70+ global languages in a 3.35B parameter model, efficient enough to run locally, even on a phone.

I will be at Nvidia GTC in March! @bharatvenki and I are gonna talk about all the systems work we do at Cohere. Come listen from me about all the custom kernel work we do for large scale LLM training on Hopper and Blackwell!! nvidia.com/gtc/session-ca…

It was truly inspiring to visit the TKMS shipyard in Kiel in my first ever visit to Germany! Building submarines is an insanely intensive engineering endeavor that spans both moving bits and atoms, in a highly sensitive secret, propreietary, and sovereign order of operations. A perfect fit for North :)

40 characters to capture the challenges, allies, and moments that define what it means to be an ML researcher. Meet the legends and get your booster pack at #NeurIPS2025. 🎴 #LabLegends

I am hiring highly skilled performance engineers for my team! You will be working on optimising pretraining for models >100B params on O(1000s) of GPUs, and hardware-aligned architecture design. We are cooking a lot of very exciting projects and I can safely say you will have a lot of fun! Link in thread. <3

Cohere is expanding in EMEA! We’re excited to announce our new Paris office as a strategic hub to strengthen our growing operations and better support customers and partners across the region. Join us building the future of enterprise AI: cohere.com/blog/paris-off…

@cohere's AI infra team is hiring software engineers in the UK & EU! If you love Kubernetes, tons of compute, and helping to build EGI: we need you! 🫵 Apply below! 👇

Introducing Command A Translate, our state-of-the-art model designed for high-quality translation tasks.

Introducing Command A Reasoning, our most advanced model for enterprise reasoning tasks.

Our secure agentic AI platform, North, is now widely available.

cohere vision model :) weights on huggingface huggingface.co/blog/CohereLab…

Introducing Command A Vision, a state-of-the-art generative model that excels across multimodal image capabilities that matter for enterprises!