#cohere

The draw for LocalLLaMA was not just another coding model, but Cohere asking the local-inference crowd to test pre-release weights first.

Europe’s sovereign AI argument just gained a balance sheet. Cohere and Aleph Alpha plan to combine, while Schwarz Group companies line up $600M (€500M) in financing to turn compliance, local hosting and frontier-model scale into one offer for governments and regulated industries.

Why it matters: inference cost is now a product constraint, not only an infrastructure problem. Cohere said its W4A8 path in vLLM is up to 58% faster on TTFT and 45% faster on TPOT versus W4A16 on Hopper.

Cohere has entered the speech stack race with Transcribe, a 2B Apache 2.0 ASR model for 14 languages. Open weights, Hugging Face distribution, and a claimed 5.42 average WER headline the release.

Cohere and Saab have signed an MOU aimed at bringing advanced AI into GlobalEye and other secure aerospace workflows. The initial scope includes data-driven mission support, maintenance tools, and on-premises information processing in secure environments.

Cohere said on March 25, 2026 that its frontier AI models now power RWS Language Weaver Pro for high-stakes enterprise and government translation workflows. RWS says the product is a 100+ billion parameter model built with Cohere, ranked first in 31 of 32 languages in its benchmarks, and outperformed DeepL and Gemini on sentence-level and paragraph-level tests.

Cohere said on March 28, 2026 that Transcribe is setting a new bar for speech recognition accuracy in real-world noise and linked users to try it. The supporting Hugging Face materials position Transcribe as an Apache 2.0, 2B-parameter ASR model for 14 languages, while a companion WebGPU demo shows the model running locally in the browser.

Cohere announced Transcribe on March 26, 2026 as an open-source speech recognition model. Cohere says the 2B Conformer-based system supports 14 languages, tops the Hugging Face Open ASR Leaderboard with 5.42 average WER, ships under Apache 2.0, and is available for download, API use, and Model Vault deployment.

In a February 20, 2026 (UTC) X post, Cohere said conversations at the India AI Impact Summit focused on responsible frontier AI scaling and language accessibility. The company tied this to Tiny Aya and New Delhi commitments.

At the India AI Summit on February 17, Cohere released Tiny Aya, a family of 3.35B open-weight multilingual models supporting 70+ languages that run offline on standard laptops, targeting global language accessibility.