- Featurecohere.com
Release Recap: Scaling Inference-Time Compute with Teams of Coding Agents
Cohere published research showing that strategic delegation in multi-agent systems can reduce costs by 70% while maintaining near-optimal performance, using a 17-agent heterogeneous team that matches …

- Contentcohere.com
Research papers
Cohere's research papers page lists its recent publications, running from May to late October 2026. Recent titles cover speculative decoding on sparse MoE models, safety and refusal for African langua…
- Contentcohere.com
Why Speculative Decoding Works Better Than Predicted on Sparse MoE Models
Cohere published a research paper, dated October 28, 2026, examining why speculative decoding, a technique that speeds up autoregressive inference by verifying multiple draft tokens in a single forwar…

- Case Studycohere.com
Ada finds the right ingredients for faster, more efficient AI agents
Cohere published a customer story on Ada, a customer service AI company that evaluated four embedding approaches across 15,000 queries. Ada uses Cohere Model Vault for offline knowledge-base indexing …

- Contentcohere.com
AFRIGUARD: Safety and Robust Refusal For African Languages
Cohere's research section published AFRIGUARD, a paper on safety and robust refusal behavior in large language models for African languages. The paper argues that safety risks are expressed and interp…

- Launchcohere.com
Compass indique la voie à suivre pour obtenir des perspectives stratégiques utiles
Cohere unveiled Compass, a new enterprise-grade search and discovery platform designed to extract contextually relevant insights from internal data sources. Compass supports multimodal and multilingua…
- Launchcohere.com
Sécurité, efficacité et mise en production rapide pour l’intelligence agentique
Cohere introduced the Command family of models, designed for real-world agentic AI applications in enterprises, emphasizing security, efficiency, and rapid deployment. Command A is positioned as the m…
- cohere.com
Traduction automatique sous votre contrôle
Cohere introduced North Small Translate, a new open-weight translation model optimized for speed, precision, and control. It achieves an 83.6 score on WMT26 benchmarks across all languages, making it …
- cohere.com
Transformez vos documents d'entreprise en données prêtes pour l'IA
Cohere introduced Parse, a new document parsing model that converts complex business documents, tables, and images into structured data for AI applications. Parse supports high-fidelity OCR, multimoda…
- cohere.com
Embed 5 est là
Cohere unveiled Embed 5, a new family of frontier embedding models that transform text and images into rich semantic representations for enterprise search, RAG, and agent retrieval. Available immediat…
- cohere.com
Optimisez votre recherche d’entreprise grâce à un classement précis des résultats
Cohere introduced Rerank 5, a new reranking model designed to improve search precision in enterprise pipelines by reducing token usage, latency, and noise. The model supports Context Focus, which extr…
- Featurecohere.com
Déploiements privés
Cohere introduced new private deployment options for its AI models, including Model Vault for isolated inference and cloud private (VPC) deployments, alongside on-premise options. These solutions ensu…
- docs.cohere.com
Cohere updates Rerank API v2 with structured data support and examples
Cohere’s Rerank API v2 now supports structured data inputs (YAML-formatted documents) alongside plain text, improving semantic search performance for structured content. The update includes updated do…
- Featuredocs.cohere.com
Cohere updates Rerank API best practices with v5.0 and v4.0 constraints
Cohere revised its Rerank API documentation to provide detailed performance constraints and chunking behavior for models v5.0, v4.0, v3.5, and v3.0. Key updates include maximum document limits (10,000…
- Launchdocs.cohere.com
Cohere launches Rerank 5 Pro and Fast models with Context Focus and calibrated scores
Cohere introduced Rerank 5 Pro and Fast, its most advanced reranking models for enterprise retrieval, featuring Context Focus for passage extraction and calibrated relevance scores. The models support…
- Launchcohere.com
Sovereign machine translation
Cohere introduced North Small Translate, a new open-weight machine translation model optimized for speed, accuracy, and cost efficiency. It achieves an 83.6 WMT26 benchmark score across all languages,…
- Contentcohere.com
Scaling Inference-Time Compute with Teams of Coding Agents
Cohere published a research paper arguing that heterogeneous multi-agent coding teams can reach near-optimal rewards at a fraction of homogeneous-team cost. The approach assigns cheap models to planni…

- cohere.com
Votre IA,selon vos règles.
Cohere introduced North 2, its latest enterprise AI platform update, featuring over 15 new capabilities including the new Embed 5 family of models, Transcribe, Parse, and Aya. The platform emphasizes …
- cohere.com
Schnell entwickeln. Kontrolle behalten.
Cohere introduced Model Vault, a fully managed SaaS inference platform offering isolated, high-performance model deployment with zero operational overhead. It supports all Cohere models, guarantees un…
- cohere.com
Behalten Sie die Kontrolle über Ihre Daten und Ihre Zukunft mit Sovereign AI
Cohere introduced North 2, featuring over 15 new features including Model Vault for isolated, on-prem or VPC-based inference, and Embed 5, a new family of frontier embedding models. The update targets…
- cohere.com
Développez rapidement. Gardez le contrôle.
Cohere introduced Model Vault, a managed SaaS platform for isolated model inference that combines API simplicity with private hosting security. It supports all Cohere models (generation, reranking, em…
- Contentcohere.com
Préparez l’avenir de votre institution financière grâce à l’IA pour les entreprises
Cohere’s financial services page highlights North 2 and new models like Embed 5, emphasizing secure, compliant, and customizable AI deployments for institutions. It positions Cohere as a solution for …
- Launchcohere.com
Modernisez les services publics, grâce à l’IA pour le lieu de travail
Cohere introduced North 2, its latest enterprise AI platform update with over 15 new features, and Embed 5, a new family of frontier embedding models. These launches target public sector use cases, en…
- cohere.com
Prenez le contrôle de vos données et de votre avenir grâce à l’IA sous votre contrôle
Cohere introduced North 2 with over 15 new features, including Embed 5 models, Parse for document analysis, and enhanced agentic capabilities. The update targets telecom operators with solutions for c…
- cohere.com
Expédiez des logiciels plus intelligents grâce à l’IA pour les entreprises
Cohere introduced North 2, its latest enterprise AI platform update, featuring over 15 new capabilities including Embed 5 for embeddings, Transcribe for speech recognition, Parse for document analysis…
- Featurecohere.com
Maximisez l’efficacité et le temps de disponibilité grâce à l’IA pour les entreprises
Cohere introduced North 2, its latest enterprise AI platform update, featuring over 15 new capabilities tailored for manufacturing. Key additions include Model Vault for isolated inference, new models…
- cohere.com
The Agentic Task Ecosystem: A Supply-Side Record of Agentic Automation
Cohere published a research paper introducing the Agentic Task Ecosystem (ATE), a tagged dataset of 694,411 public MCP tools mapped against 178 occupations. The paper finds only 2.6% of tools cleanly …

- Featuredocs.cohere.com
Cohere launches Model Vault for dedicated, single-tenant model serving
Cohere introduced Model Vault, a managed, single-tenant environment for deploying and serving its models with dedicated infrastructure, eliminating shared-resource noise and rate limits. The offering …
- docs.cohere.com
Cohere releases North Small Translate, an open-weights translation MoE model
Cohere launched North Small Translate, an open-weights mixture-of-experts model optimized for machine translation across 50+ languages. The model is available via the free-tier Chat V2 API and as open…
- docs.cohere.com
Cohere launches Parse v5.0 for document intelligence workflows
Cohere released Parse (model ID: parse-v5.0), a 2.3B-parameter multimodal model that converts complex documents into structured Markdown, including text, tables, forms, images, and layout metadata. Av…
- docs.cohere.com
Cohere launches open-source Arabic speech-to-text model Transcribe Arabic
Cohere released Cohere Transcribe Arabic, an open-source speech recognition model fine-tuned for Arabic audio, including regional dialects. It is the most accurate open-source Arabic ASR model availab…
- Launchdocs.cohere.com
Cohere launches Transcribe Arabic, an open-source ASR model optimized for Arabic dialects
Cohere introduced Cohere Transcribe Arabic, a 2B-parameter open-source automatic speech recognition model fine-tuned for Arabic audio inputs, supporting multidialectal and code-switching scenarios. Th…
- docs.cohere.com
Cohere launches North Mini Code, its first open-source agentic coding model
Cohere released North Mini Code, a 30B/3B parameter Mixture of Experts model designed for agentic coding, with a 256K input and 64K output context length. The model is available via the Chat V2 API, a…
- docs.cohere.com
Cohere launches Embed 5 Pro and Embed 5 Fast models
Cohere introduced Embed 5, its most advanced embeddings family, featuring two variants: Embed 5 Pro for highest retrieval quality and Embed 5 Fast for low-latency throughput. The models support multim…
- docs.cohere.com
North Small Translate
Cohere introduced North Small Translate, a 218B-parameter mixture-of-experts model designed for machine translation with 25B active parameters and a 16K context length. The model is free to use until …
- Launchdocs.cohere.com
North Mini Code
Cohere introduced North Mini Code, a 30B total/3B active parameter Mixture of Experts model optimized for agentic coding tasks. It is the first in the North family, released under Apache 2.0, and desi…
- Launchdocs.cohere.com
Cohere launches Command A+ MoE model with 25B active parameters
Cohere introduced Command A+, its first Mixture of Experts (MoE) model with 25B active and 218B total parameters, optimized for agentic, reasoning, vision, and multilingual tasks. The model supports 4…
- docs.cohere.com
Cohere deprecates multiple models, endpoints, and fine-tuning capabilities
Cohere announced the deprecation of several models, endpoints, and fine-tuning capabilities, including older command models, embeddings, rerank v2.0, and managed connectors. Shutdown dates range from …
- docs.cohere.com
Cohere updates Rerank models with v4.0 variants and expanded multilingual support
Cohere introduced two new Rerank model variants—`rerank-v4.0-pro` for high-quality multilingual re-ranking and `rerank-v4.0-fast` for low-latency use cases—alongside existing models like `rerank-v3.5`…
- docs.cohere.com
Cohere launches Embed 5, North Small Translate, Parse, Transcribe Arabic, North Mini Code, and Command A+
Cohere released six new models: Embed 5 (two variants for retrieval), North Small Translate (open-weights translation), Parse (document parsing), Transcribe Arabic (speech-to-text), North Mini Code (a…
- Launchaws.amazon.com
Cohere Parse v5
Cohere released Parse v5, a document vision parser optimized for high-throughput AI workloads, now available on AWS Marketplace. The model converts multilingual document images into structured content…
Suivez Cohere en pilote automatique
- · Brief IA hebdomadaire : résumé narratif de ce qui a été livré, chaque lundi 9 h
- · Alertes par e-mail ou Slack, ou discutez avec l'archive depuis votre tableau de bord
- · Ajoutez Cohere + jusqu'à 2 autres concurrents gratuitement, sans carte bancaire