Cohere Release Notes

North Platform v1.14.9 Release

Version Information

Platform Version: v1.14.9 (previous: v1.14.8)

North App Version: v0.350.10 (previous: v0.350.10)

Compass Chart Version: v5.7.1 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Bug fixes

Prevents deployment conflicts when horizontal pod autoscaling is enabled by allowing the autoscaler to control replica counts.

North Platform v1.13.16 Release

Version Information

Platform Version: v1.13.16 (previous: v1.13.15)

North App Version: v0.329.17 (previous: v0.329.17)

Compass Chart Version: v5.7.1 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Prevents deployment conflicts when horizontal pod autoscaling is enabled by allowing the autoscaler to control replica counts.

Original source

North 2 is Cohere’s biggest upgrade yet — unifying enterprise‑grade security, intelligence, cost governance, and full‑stack control into a single agentic AI platform that runs wherever your data lives.

Key takeaways

• New agent harness: North 2 has a redesigned orchestration system enabling more sophisticated, reliable, multi-step automations and agents across the enterprise.
• Increased intelligence: Build smarter workflows and richer agent chat experiences using memory, libraries, skills, artifact, and app creation.
• Model agnostic: Whether you’re using Cohere’s models — or bringing your own — use the models that support your team’s best work.
• AI sovereignty: You’ll get full control over where and how AI runs, with flexible cloud, on‑premises, and air‑gapped deployments.
• Enterprise governance: North admin adds granular cost controls, rate limits, user quotas, and org-wide caps for predictable, governed AI operations.

For the past few years, enterprises have been racing to deploy agentic AI, and they’ve paid for it in complexity, cost, and compromise. Every option demanded a sacrifice: the platform that's secure enough wasn’t smart enough; the one that's smart enough wasn’t controllable enough; the one you could govern became too expensive to run at scale.

North 2 is where that tradeoff ends.

Over the last year, Cohere has rolled out North across high-stakes industries, including finance, healthcare, telecommunications, manufacturing, energy, and the public sector, investing deeply in large-scale production deployments. This has shaped North 2 into a full-stack agentic platform built for organizations that can't afford to compromise on security, intelligence, cost, or control. Enterprises choose North to ensure the journey to AI ROI is more predictable, transparent, and fully owned by the enterprise.

Here's what that looks like in practice.

Intelligence that scales with your enterprise

North works with your knowledge, and now you can teach it more about what success means within your organization.

Users can now create reusable agents and automations that can be shared across the organization, eliminating the need to rebuild similar workflows repeatedly. Prototype documents or apps directly in chat, or connect North to your favorite existing tools.

At the core of North 2 is an advanced agent orchestration system that manages complex, multi-step processes independently. Build agents and automations with simple prompts, and automate your most tedious workflows — helping keep humans-in-the-loop where it matters.

Core features included in North 2

• Skills: Reusable capabilities agents can call on, so no employee wastes time rebuilding the same logic.
• Libraries: Shared knowledge and assets that the whole org can draw from.
• Memory: Agents that keep context across sessions, instead of starting cold every time.
• Applications: Prototype and ship decks, dashboards, documents, and lightweight apps from natural language.
• Automations: Pre-built templates for common processes, a drag-and-drop visual workflow builder, real-time performance monitoring, and integrations with the tools you already use.
• Connectors: Connect North to your data, with new connectors including enterprise connectors to Slack, SharePoint, OneDrive, Microsoft Outlook, Microsoft Exchange, Jira, Linear, Notion, and GitHub, plus planned financial data providers PitchBook, Crunchbase, Daloopa, FiscalAI, S&P Global, FactSet, among others.

LG CNS leverages North to bring AI to customers at scale across South Korea to unlock real business value.

“North gives enterprises a secure, practical foundation for turning agentic AI into real business value. Our experience using North within LG CNS and bringing it to customers across Korea has reinforced our confidence in its ability to scale intelligent automation across the enterprise.”
— Yohan Jin, Head of AI Center, LG CNS

Enterprise-first security

Agentic AI changes the risk calculus for every enterprise security team. With agents that touch internal tools, and models that risk exposure to prompt injection, it has never been more important to safeguard data.

North was built for that reality. It deploys privately on-prem, in the cloud, or fully air-gapped, and North Admin lets teams enforce guardrails at the individual agent level, so rolling agentic AI out to an entire workforce doesn't mean rolling out risk alongside it. Your knowledge graph gets smarter while your knowledge remains safe, never at risk of leaving the premises.

Core features included in North 2

• Deployment flexibility: Select the deployment environment that best suits your security needs (self-hosted, VPC, hybrid, or on-premises).
• Exceeding global security standards: with SOC 2 Type 2, ISO 27001, and ISO 42001 certifications.
• Guardrails: Get guardrails to filter content, improve safety, and validate model responses.
• Autonomy policies: Build agents designed to only take actions they are authorized to take, and seek human oversight for critical decisions or actions.
• Rigorous security: Proactively identify and mitigate potential threats through continual security testing, including red-teaming exercises and third-party vulnerability scanning.
• System observability: Gain visibility into North's operations with easy-to-use tools for monitoring service performance and detailed logs of all changes.

To highlight another example, Bell Cyber is partnering with North to bring government‑grade security to agentic AI.

At Bell Cyber, we see every day how quickly the threat landscape is evolving, and how much security and data sovereignty matter to the organizations we protect. North lets us bring agentic AI into our security operations with the control, privacy, and sovereignty requirements our customers demand built in from the start.
That's important not only for strengthening how we defend Bell and our customers, but for helping Canadian businesses and governments adopt agentic AI securely and with confidence.
— Jawed Ahmad, Chief Technology and AI Officer, Bell Cyber

Granular controls, improved tokenomics

North Admin delivers enterprise-grade controls. Admins set roles and permissions, and control which models get used where and by who. Additionally, admins can monitor and manage activity in real time with detailed analytics, and get granular details or see a systems-level overview of AI investment usage, including token spend and limits.

Before limits are set, thresholds for alerts can notify teams of risk. Whether your organization measures adoption or model consumption, Cohere provides the tooling to measure AI ROI.

Core features included in North 2

• North Admin: Configure visibility to your desired level, including down to the user and agent level, so enterprises know exactly how their AI is being used, and can guardrail it accordingly.
• Granular access controls and permissions: Integrate with your existing identity and access management systems, and get enhanced authentication and granular admin controls.
• Flow control: Monitor and control token spend and limiting by defining consumption tiers based on requests and token rates for users and groups.
• Efficient token consumption: When using NVIDIA hardware with Cohere models, get increased tokens per second per node, effectively decreasing spend.
• Bring your organization’s existing models: North is model agnostic, use our generative model offerings or bring your favorites.

Additionally, our collaboration with NVIDIA finds that North generates more insights, for fewer tokens, with Cohere models trained and deployed on NVIDIA Blackwell and Hopper. That means fewer spent tokens, without sacrificing model intelligence. See more on our models here.

“Organizations need control over where their AI runs and how their data is secured,” said Kari Briski, Vice President of Generative AI at NVIDIA. “NVIDIA Blackwell and Hopper accelerate inference for Cohere models, helping North 2.0 deliver more tokens for less.

Find your organization’s true North

For organizations that demand rigor in their AI investments, North 2 leads with production-ready, full-stack AI solutions. With North, get end-to-end, customizable, capable, and secure tooling that integrates with existing systems and helps protect sensitive data — that point teams on the path to true AI for empowerment.

North was built to be owned, and it is increasingly built to be specialized. With North, even the most regulated industries can adopt agentic AI confidently, knowing their deployments align with their operational, compliance, and governance needs.

Contact sales to book a demo today.

Original source

North Platform v1.15.4 Release

Version Information

Platform Version: v1.15.4 (previous: v1.15.3)

North App Version: v0.370.9 (previous: v0.370.8)

Compass Chart Version: v5.17.0 (previous: v5.17.0)

Compass App Version: v0.93.1 (previous: v0.93.0)

Bug fixes

Restores the Migrate to libraries option in Agent Builder so users can convert an agent's existing file selections into libraries.

Ensures viewers of shared Microsoft libraries only see connector files they have permission to access.

Updates Dex connector validation to support the latest token exchange settings.

Prevents authentication cookies from being sent over unsecured HTTP connections.

Adds support for ingesting XML files identified with the text/xml media type.

Original source

Key takeaways

Our most powerful embedding models yet, now available in Pro and Fast tiers.

• State-of-the-art enterprise retrieval: Embed 5 Pro achieves the highest average score of any model we tested - particularly across financial datasets, parsed PDFs, and visually rich documents.
• A new Fast tier: Embed 5 Fast brings strong retrieval quality to latency - and cost-sensitive workloads, at $0.08 per million tokens.
• One index, two models: Pro and Fast share an embedding space, so teams can index with Pro and query with either model without re-indexing.
• Built for complex enterprise data: Embed 5 supports multimodal inputs and retrieval, 100+ languages, and a 128K-token context window for longer documents.
• More efficient at scale: Matryoshka representations and lower-precision outputs reduce vector storage and search costs, while quantized weights lower serving requirements for private deployments.

Today, we're releasing Embed 5, a new family of embeddings models at the frontier of high-quality enterprise retrieval.

Embed 5 delivers stronger retrieval across complex enterprise data while giving teams more control over latency, cost, and deployment. Embed 5 Pro is optimized for maximum quality across multimodal, multilingual, financial, code, and parsed-document retrieval. Embed 5 Fast brings highly competitive performance to latency- and cost-sensitive workloads. Both tiers share a single embedding space, so teams can index with Pro and query with either model without rebuilding the index.

Embed 5 establishes the retrieval foundation for search, RAG, and agentic workflows, surfacing more relevant context while filtering out noise before it reaches expensive generative models. Use Embed to improve answer quality and user experience while helping keep downstream inference costs under control.

Embed 5 is generally available today on the Cohere API and Model Vault, Microsoft Foundry, and Amazon SageMaker. Pricing is $0.12 per million tokens for Pro and $0.08 per million tokens for Fast.

Snapshot

Capability Embed 5 Pro Embed 5 Fast Best for Maximum retrieval quality; offline indexing; complex enterprise corpora Interactive search; high-volume RAG; agentic retrieval Context length 128K tokens 128K tokens Inputs Text, images, fused text + image Text, images, fused text + image Languages 100+ 100+ Output dimensions 2048, 1536, 1024, 768, 512, 256 2048, 1536, 1024, 768, 512, 256 Embedding formats float, int8, binary float, int8, binary Matryoshka embeddings Yes Yes Shared embedding space Yes Yes Supports self-hosting Yes Yes Pricing $0.12 / 1M tokens (text) $0.40 / 1M tokens (image) $0.08 / 1M tokens (text) $0.40 / 1M tokens (image)

Performance

Embed 5 Pro delivers our strongest retrieval performance to date. It achieves the highest average score of any model we tested across ViDoRe V3, financial documents, parsed PDFs, image retrieval, and across key business languages.

Embed 5 is also the first model family evaluated with RCP-nDCG@10, our latest retrieval methodology. Instead of scoring only against a limited set of fixed labels, it evaluates retrieved documents against query-specific relevance criteria, capturing relevant results and giving a fuller view of performance on your own corpus.

Enterprise documents

Embed 5 excels with visually rich documents where meaning lives in tables, charts, diagrams, and layout - not just text. On ViDoRe V3, which features documents sampled across key enterprise domains including financial filings, technical manuals, regulatory material, government reports, textbooks, and lectures, Embed 5 Pro averages 85.8 - an impressive 8.8 gain from Embed 4.

That puts it ahead of Voyage 4 Large (83.7), Gemini Embedding 2 (83.2), and OpenAI text-embedding-3-large (75.5). Pro leads five of the eight domains outright and ties Voyage 4 Large on energy, with its largest gains over Embed 4 on HR (+11.4) and industrial (+10.3). Embed 5 Fast averages 84.5, ahead of both Gemini Embedding 2 and Voyage 4 Large.

Finance

Embed 5 Pro establishes itself as the leading embeddings model for financial document retrieval.

Pro ranks first on three leading public financial benchmarks, with Fast second on each despite being considerably smaller than its peers: FinanceBench (80.1 Pro, 80.0 Fast), FinQA (90.0, 88.8), and ViDoRe V3 Finance (85.0, 83.9).

Across these, Pro averages 3.3 points higher than the next non-Cohere competitor, Gemini Embedding 2. Compared with OpenAI text-embedding-3-large, the lead grows to 21.4 points on FinanceBench.

Multimodal

Parsed PDFs

Most enterprise search pipelines still convert PDFs to text before embedding them, but that process can strip away structure. Tables lose row and column relationships, multi-column layouts can scramble reading order, repeated headers add noise, and charts often disappear entirely. That makes parsed-document retrieval a harder test than clean-text benchmarks suggest.

Our parsed-document suite spans service documentation, corporate reports, SEC filings, product manuals, and privacy policies. Embed 5 Pro achieves the highest average across the suite at 84.8, ahead of Voyage 4 Large at 83.6, Embed 5 Fast at 83.4, Gemini Embedding 2 at 80.8, and Embed 4 at 78.6.

Page-image and fused text-image documents

Some documents are better represented visually. Scanned pages, slide decks, schematics, and charts contain information that text extraction may miss. Embed 5 can embed page images directly (page-image), or combine an image with its metadata into a single vector (fused text-image).

On fused text-image corpora, Embed 5 Pro averages 82.3 across five datasets, ahead of Embed 5 Fast at 81.2 and Gemini Embedding 2 at 61.3. Pro outperforms Gemini Embedding 2 on every dataset in the suite. Page image retrieval is also robust: Embed 5 Pro continues to lead on financial datasets, averaging 77.0 from five datasets, ahead of Embed 5 Fast (73.2), Embed 4 (71.1), Voyage Multimodal 3.5 (70.1), and Gemini Embedding 2 (56.7).

Multilingual

Embed 5 is trained on more than 100 languages, with particular focus on the languages most used by our global customer base.

Across German, French, Spanish, Italian, and Russian, Embed 5 Pro achieves the highest average of the models we tested: 77, compared with 76 for Voyage 4 Large, and 73 for Gemini Embedding 2. It improves on Embed 4 by around 7 points on average, with the largest gains in Russian (+9) and Italian (+7).

The table below covers ten further languages where Embed 5 has made important strides against Embed 4. Pro’s largest gains are in middle eastern and subcontinent languages, notably Farsi (+13), Telugu (+12), and Hindi (+12).

Meet Embed 5 Fast

Embed 5 Fast is a lighter weight model built for latency-sensitive, high-volume retrieval. It costs a third less than Pro while retaining the same 128K-token context, multimodal inputs, multilingual coverage, and multiple compressed output formats.

That matters most on the query path, where embedding latency is paid on every search - and multiplied in agentic workflows that may issue dozens of searches per task. Fast’s smaller footprint also lowers serving costs in private deployments and speeds large ingestion and re-indexing jobs.

For document throughput - a closer proxy for indexing efficiency - Fast is consistently more efficient, delivering an average of 2.4× higher throughput than Pro across context sizes.

Performance

Fast raises the bar for compact embedding models. On ViDoRe V3, it leads Voyage 4 Nano by almost seven points and Jina Embeddings v5 Text Small, Perplexity, and Microsoft’s Harrier 0.6B by ten or more. It outperforms Qwen3-VL-Embedding-2B, despite being roughly half the size, by about 20 points. Its average also exceeds Gemini Embedding 2 and Voyage 4 Large on ViDoRe V3 and on financial retrieval. On parsed PDFs it exceeds Gemini Embedding 2 (83.4 vs 80.8) and trails Voyage 4 Large (83.6).

Two models, one embedding space

Pro and Fast share a single embedding space, so vectors from either model can be compared directly. We tested every corpus/query pairing across 40 development datasets spanning text, image, fused, and parsed-document retrieval.

That shared space lets teams choose each tier independently: documents can be indexed with Pro for maximum quality, while queries use Fast for lower latency and cost—without rebuilding the index. The cross-model combinations remain close to the same-model baselines (averaging just 1.6% and 2.7% losses for Fast and Pro queries, respectively), with no dataset showing a major failure.

For many customers, we recommend the following deployment pattern: index with Pro, query with Fast. It captures much of the quality gain of an all-Pro system while keeping Fast’s latency and cost during request.

Vector storage

At enterprise scale, the vector index can cost more to operate than the model that generates it. Embed 5 supports Matryoshka representation learning and lower-precision outputs, letting teams shrink vectors and finely control the tradeoff between quality, storage, and search cost. These savings can be substantial - a 2,048-dimensional float32 vector requires 8 KB; a 1,024-dimensional int8 vector uses 1 KB; and a 256-dimensional binary vector just 32 bytes—a 256x reduction. Across 100 million chunks, that cuts raw vector storage from roughly 819 GB to 3.2 GB.

Importantly, int8 retains near-full-precision retrieval quality in both Embed 5 Pro and Fast. For most deployments, we recommend 1,024-dimensional int8 vectors as the ideal performance-efficiency point. Binary offers the smallest footprint, with some accuracy tradeoff, and is well suited to fast first-pass retrieval before higher-precision reranking.

Getting started

Deploy Embed 5 Pro and Embed 5 Fast through the Cohere API, Model Vault, Microsoft Foundry (Pro, Fast), and Amazon SageMaker (Pro, Fast), or use Embed 5 directly within North. For private deployments in your own VPC or on-premises, both models can be served with vLLM. Batch embedding is available for large-scale ingestion.

Build with the tools you already use. Embed 5 fits into existing retrieval stacks, with integrations across frameworks and vector databases including LangChain, Haystack, Weaviate, Qdrant, Pinecone, Elasticsearch, MongoDB, Redis, Milvus, and OpenSearch.

Start by creating an API key, then use the code snippets below to quickly make your first query.

What else

Meet the team behind Embed 5. Join us on X on October 8 to hear from our search and embeddings leadership about Embed 5, Parse 5, and what else we’ve been preparing behind the scenes.

Also, Compass Cloud, our managed search and retrieval platform, is now in private beta. Request access to try it on your own retrieval and agentic workloads.

Key contributors

Samarth Bhargav, Fabian Schmidt, Clifton Poth, Arthur Maciejewicz, Florian Schneider, David Rau, Dennis Zhao, Timothy Ang, Nils Reimers, Carlos Lassance.

Footnotes

1. RCP-nDCG@10 requires evaluating embedding models in a two-stage retrieval setup, using their similarity scores to reorder a fixed candidate set. Scores therefore reflect reranking quality rather than first-stage retrieval performance, which we thoroughly evaluate elsewhere against nDCG and Recall.

2. The annotations and code needed to evaluate Vidore V3 with RCP-nDCG are available here.

3. Both sides must use the same output dimension. Compatibility also holds with Matryoshka truncation and int8 quantization, so the same pattern works with compressed indexes.

Original source

We're pleased to announce the release of Embed 5, Cohere's most powerful embeddings family yet.

Embed 5 delivers frontier retrieval quality on complex enterprise data, with major gains over Embed 4 on visually rich documents, financial filings, parsed PDFs, code, and multilingual retrieval.

Key features

• Two model variants available:embed-v5.0-pro: Optimized for the highest retrieval quality, particularly for offline indexing and quality-critical retrieval
• embed-v5.0-fast: Optimized for low latency and high throughput, particularly for interactive search, agent loops, and high-volume query traffic
• Shared embedding space: Pro and Fast share an embedding space, so a corpus indexed with one model can be queried with the other. We recommend indexing with Pro and querying with Fast.
• Multimodal inputs: Embed text, images, and mixed text-and-image inputs (e.g. PDF pages) in a single vector
• Multilingual support: Supports over 100 languages
• Extended context length: 128k token context window
• Flexible storage: Matryoshka embeddings in the following dimensions: [256, 512, 768, 1024, 1536, 2048], with float, int8, and binary output types

Availability

Embed 5 is available through the Embed API, as well as Microsoft Foundry (Pro, Fast) and Amazon SageMaker (Pro, Fast).

For single-tenant deployment, Embed 5 is also available in Cohere's Model Vault.

For more details, see the model documentation.

Original source

North Platform v1.15.3 Release

Version Information

Platform Version: v1.15.3 (previous: v1.15.2)

North App Version: v0.370.8 (previous: v0.370.8)

Compass Chart Version: v5.17.0 (previous: v5.14.0)

Compass App Version: v0.93.0 (previous: v0.93.0)

Compass features

Adds an optional access check so that search results from Atlas only include documents the current user is allowed to open.

Adds optional configuration for how often Google Drive content resyncs.

Bug fixes

Fixes SharePoint Sites.Selected settings so an admin can choose the permission mode and limit syncing to an allowlist of sites.

Improves visual PDF parsing by applying a dedicated page-size limit when the visual parser prepares page images.

Fixes platform status checks that reported a healthy visual parsing model as unreachable.

Prevents temporary files from an interrupted parse from lingering and using up storage unnecessarily.

North Platform v1.13.15 Release

Version Information

Platform Version: v1.13.15 (previous: v1.13.14)

North App Version: v0.329.17 (previous: v0.329.16)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Fixes a server error when registering or updating an MCP server that exposes a tool and a prompt with the same name. Existing tools, prompts, and permissions keep their current names.

Original source

Compass is Cohere’s retrieval platform for developers building AI applications with their enterprise data

It surfaces the most relevant information from your company’s corpus for use in retrieval-augmented generation (RAG), search, and agentic workflows. Now, Compass is entering private beta as a managed offering - Compass Cloud.

Compass was built for builders. It provides the retrieval foundation that developers can freely configure around their data and workflow requirements. Instead of assembling and operating the stack themselves, teams access Compass through its APIs, MCP server, or Python SDK to shape their desired end user experience.

Until now, Compass has principally powered retrieval for North, Cohere's enterprise agent workspace, including its document libraries and MCP ecosystem. We have also shipped Compass into highly secure, self-hosted environments for partners in regulated industries whose workloads cannot be offloaded to SaaS.

Customer demand for a managed option has been clear and consistent: teams want Compass' best-in-class retrieval capabilities, but many do not want the operational overhead that comes with self-hosting.

We’ve listened to those calls. Compass Cloud expands Compass to a broader market. It lets Cohere manage the full pipeline and model inference, so that our customers can focus even more on building. In parallel, self-hosted deployments remain available for privacy-constrained projects.

We're working with a limited number of enterprise teams as beta partners. Interested? Request access.

Compass is solving open problems in enterprise search

Search and retrieval have improved, but enterprise performance is no longer defined by relevance and latency for a single query. As retrieval becomes key supporting infrastructure for generative AI and agents, three developments are changing the requirements:

1. Token economics: Every irrelevant result passed to a model consumes tokens and occupies limited context space. More precise retrieval creates smaller, higher-quality inputs, reducing inference costs and cutting the time needed to complete a task. Retrieval is one of the most effective cost levers available to businesses today.

2. Agentic access patterns: Agents may issue dozens of queries while completing one task, reformulating requests and traversing multiple sources. In these multi-hop loops, latency accumulates, relevance can drift, and permissions must be enforced at every step. Retrieval must therefore perform reliably across sequences of machine-generated queries, not only single-shot searches.

3. A fragmented retrieval stack: Production pipelines often combine separate systems for ingestion, indexing, reranking, access control, and orchestration. Different middleware and sources of truth leave teams spending substantial effort on integration rather than retrieval quality.

Compass exists to address each of these: 1) by providing relevant, governed context for generative workloads; 2) supporting both agentic and conventional search applications; and 3) consolidating the core retrieval stack into an integrated enterprise platform.

Full-stack search and retrieval

Compass packages document processing and retrieval into one configurable service. Teams can access a single interface instead of integrating and operating separate services.

Cohere Compass architecture: a governed enterprise retrieval stack connecting data sources to front-end applications through parsing, embedding, hybrid search, and reranking.

Connect: Access out-of-the-box connectors for your file sharing and cloud storage workspaces, such as SharePoint, OneDrive, and Google Drive. Quickly access the content you need with Compass’ near-universal data compatibility — multilingual, multimodal, and file format-agnostic.

Parse: Turn complex documents into searchable, structured data. Compass transforms multimodal enterprise files into AI-ready content, applying the right parsing strategy to each document and using vision processing only where it adds value, reducing unnecessary model usage.

Embed: Capture meaning and exact terminology. Compass generates dense and sparse representations together, so search can match both semantic intent and domain-specific language across text and multimodal content.

Index: Keep your source files, parsed content, and embeddings as separate records, so a new embedding model can be adopted without crawling and uploading the same content again. At search time, they sit in one index with their metadata, reducing the need to keep each system in sync.

Retrieve: Combine search strategies in one request. Semantic, sparse, and keyword search can run independently or together, balancing recall and precision. Permissions are enforced during retrieval rather than left to the application.

Rerank: Send stronger evidence to the model. Cohere’s best-in-class reranker identifies the most relevant passages from a broad candidate set, reducing irrelevant context and the tokens required for generation.

Govern: Enforce multi-tenant access control and document-level permissions during retrieval, so applications don’t have to filter results themselves. In Compass, retention policies expire content automatically and prevent deleted documents from being resynced.

Premium retrieval accuracy

The core objective of Compass is simple: improve the relevance of information surfaced for enterprise knowledge applications, be that a RAG pipeline or an autonomous agent. That quality is rooted in Cohere's best-in-class search and document-processing models together with Compass's hybrid search: lexical, sparse, and dense retrieval, then rerank.

The figure below is one instance of the accuracy gain over traditional or standalone search infrastructure on a representative financial-industry RAG workload: embed a query, retrieve presentation materials from an index, and score the top results. On High Finance, a Cohere-built investment-banking benchmark, Compass achieved a 14-16 point improvement on Azure Search (from 64.8 to 81.1). A gap of this size can be the difference between an unsatisfactory answer and a great one for the end user.

Build on your terms

Users can upgrade their search performance with Compass Cloud in two key ways, depending on how they want to integrate retrieval into their application:

Compass API: Best for teams building custom RAG and search applications. Use Compass APIs to integrate directly with the models and retrieval components that power Compass, giving developers more control over how content is represented, retrieved, ranked, and incorporated into their application.

Model Context Protocol (MCP): Best for agentic retrieval. Compass ships a dedicated, independently versioned MCP server that exposes retrieval as reusable tools through an open standard. MCP-compatible clients can discover and call those tools directly, making Compass portable across agent frameworks without a bespoke integration for each one.

Agentic retrieval is an emerging approach at the intersection of large language models and information retrieval. Rather than relying on a single query-and-response retrieval step, an agent can decide what to look for, search for relevant information, evaluate the results, and refine its search before answering.

This is especially useful for tasks that cannot be resolved with a single lookup. With the Compass MCP server, agents can progressively narrow the retrieval corpus, reducing unnecessary context and making complex retrieval tasks faster and more token-efficient.

Get early access

We’re opening this private beta for teams building retrieval-heavy and agentic applications.

Beta participants get hands-on support from the engineering team, early access to the cloud deployment, and direct influence on the roadmap. Request access.

We’re also running a live session on October 8th on X with Cohere’s engineering and product leadership on where enterprise search goes next.

Original source

North Platform v1.15.2 Release

Version Information

Platform Version: v1.15.2 (previous: v1.15.1)

North App Version: v0.370.8 (previous: v0.370.7)

Compass Chart Version: v5.14.0 (previous: v5.14.0)

Compass App Version: v0.93.0 (previous: v0.93.0)

Bug fixes

Sorts numeric table columns numerically instead of alphabetically.

Addresses an issue where images could not be uploaded to SharePoint libraries.

Fixes a security issue impacting document generation that could allow unauthenticated requests under certain conditions.

North Platform v1.14.8 Release

Version Information

Platform Version: v1.14.8 (previous: v1.14.7)

North App Version: v0.350.10 (previous: v0.350.9)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Bug fixes

Fixes a security issue impacting document generation that could allow unauthenticated requests under certain conditions.

North Platform v1.13.14 Release

Version Information

Platform Version: v1.13.14 (previous: v1.13.13)

North App Version: v0.329.16 (previous: v0.329.15)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Fixes a security issue impacting document generation that could allow unauthenticated requests under certain conditions.

Original source

North Platform v1.15.1 Release

Version Information

Platform Version: v1.15.1 (previous: v1.15.0)

North App Version: v0.370.7 (previous: v0.370.6)

Compass Chart Version: v5.14.0 (previous: v5.14.0)

Compass App Version: v0.93.0 (previous: v0.93.0)

Bug fixes

Fixes Admin Analytics v2 trend charts that previously shortened the selected date range when some time periods had no data. Now, empty time periods stay on the axis as gaps instead of disappearing or reading as zero usage.

Updates backend dependencies to address known CVEs.

North Platform v1.14.7 Release

Version Information

Platform Version: v1.14.7 (previous: v1.14.6)

North App Version: v0.350.9 (previous: v0.350.9)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Features

Adds optional Helm values so you can inject extra init containers, volumes, and volume mounts on the Admin frontend Deployment, matching the existing toolkit-frontend overlay hooks.

Original source

North Platform v1.15.0 Release

Version Information

Platform Version: v1.15.0 (previous: v1.14.6)

North App Version: v0.370.6 (previous: v0.350.9)

Compass Chart Version: v5.14.0 (previous: v5.7.0)

Compass App Version: v0.93.0 (previous: v0.80.0)

Agents

[Alpha]Build agents with anAI assistant by describing what you want in natural language. Instruct the assistant to set up your agent for your use case or ask it questions about the configuration.

The agent builder has been redesigned to become a more consolidated and user-friendly experience that allows for more space to write formatted agent instructions and makes it easier to see enabled tools and capabilities.

[Beta]Automatically drafted version notes are generated based on the changes you made when you publish a custom agent. You can revise the notes before publishing.

Duplicating an agent now also copies its evaluation tasks and test cases to ease the process of creating thoroughly evaluated agents.

Automations

[Beta]Redesigned Automations list.

The Automations page has been made more powerful and easy to navigate by consolidating it to just two tabs (Discovery and Runs) and offering many more filtering options.

[Beta]Automatically drafted version notes are generated based on the changes you made when you publish an automation. You can revise the notes before publishing.

[Beta]Describe an automation schedule in plain language and North fills in the schedule for you to review before you save.

Chat

Chat tools and capabilities are now consolidated into a single menu that is easy to use and search. From the plus icon, you can upload files, add libraries, switch modes (e.g., deep research), and manage tools from a single, streamlined location.

An agent selector on the homepage lets you choose which agent to chat with before you send, making the chat more versatile. You can configure the default selection in user settings.

Recent chats are now easier to navigate in the sidebar, showing conversation history as a single, chronological list of chats with all of your agents.

Chat and sidebar elements now load progressively as they are available, resulting in a smoother experience when navigating North with a slower connection.

[Alpha]Session compaction replaces older conversation history with an AI-generated summary as a chat session approaches its context limit, letting long conversations continue instead of hard-truncating history or hitting a context error. Contact your Cohere representative to enable it.

Deep research now runs only when the agent decides a prompt calls for its full multi-step research process, the same way document mode's canvas editor already opens only when the agent decides a prompt calls for one. If unsure, the agent may ask the user.

My files

Expanded supported file types. My files and local library uploads now match Compass for accepted formats, including more spreadsheet types (e.g., TSV, ODS, and macro-enabled Excel) and additional document formats such as JSON, RTF, EPUB, and OpenDocument.

Images attached in chat are now also saved to My files, the same as any other attachment, so you can reuse them in other conversations or file workflows—understanding the image in the original chat is unchanged.

Known limitation: images aren't indexed for Smart Search, so an agent won't find one just because it's sitting in your files. Reference the image directly—highlight it in chat or attach it again—to have an agent look at its contents.

Tables

[Alpha] Review mode shows every cell's review state—Reviewed, Unreviewed, or Out of date—at a glance.

[Alpha] Filtering by review state narrows the table to rows with a cell in a specific review state—Reviewed, Unreviewed, or Out of date—while in Review mode, so you can focus on, for example, just the cells that still need review.

[Alpha] Table cell citations now open in a dedicated split-screen viewer that highlights the exact source page next to the cell's answer, so you can verify a grounded response without leaving the table.

Admin

Tool plan editing is removed. As announced in v1.12.2, the following configuration found under Chat Settings has been fully removed: Allow users to edit the tool plan during a chat session.

[Alpha] Data retention now uses a single global toggle, off by default. The retention period is now measured from a conversation's creation date rather than its last activity. Its soft-delete timeout can now be set to 0 days, skipping any recoverable window before permanent deletion.

[Alpha] SCIM provisioning lets North synchronize users and groups with your identity provider across their full lifecycle—including deletions and updates.

[Alpha] New Analytics dashboard powered by ClickHouse adds deeper drill-downs across adoption, agents, automations, and tools.

[Alpha] Sandbox observability gives platform admins eBPF-based runtime visibility into sandbox pods—process execution, file access, and network activity—with an optional web interface for browsing captured sessions.

Two new system roles for compliance workflows—Compliance Requester and Compliance Approver—make it easier for admins to provision compliance staff with access. Organization admins no longer automatically get Can approve compliance requests, nor can they grant either permission to themselves or to a group they belong to—someone else must grant it.

New Library Access Manager system role lets admins grant trusted users control over organization-wide library visibility and membership without granting them full organization admin rights.

Continues the design refresh and migration to React for the Agents, Automations, Analytics (default), Compliance, Feedback dialogs, Audits, and Notification banners pages.

Libraries

Attach a library to a deep research session from the composer so North can search that library's contents during the research run.

A new Libraries page in the North admin panel lets admins browse libraries across the organization, filter by source (e.g., My files), and manage their visibility and user access from one place.

Manage library ownership for a deleted user's shared libraries by assigning them to a new owner, which allows the new owner to manage and modify the libraries going forward. Note that ownership transfer is not possible for Microsoft connector libraries (e.g., SharePoint).

Graceful Delete Settings are now applied to a deleted user's libraries in addition to other artifacts. Note that shared Microsoft connector libraries are always hard-deleted after the grace period because they are not transferable.

Local file library size limits can be configured by admins to set a storage cap per library.

Guardrails

New Guardrails gRPC API lets trusted applications call North guardrails directly over gRPC to scan prompts, model responses, and tool output.

Guardrails deployment now runs through North's own cohere-guardrails subchart instead of HiddenLayer's bundled externalApi. If you're upgrading an existing deployment that used the old path, set externalApi.enabled: false on the HiddenLayer chart and enable cohere-guardrails on cohere-eno. The AIDR scanning engine itself is unaffected—only its bundled guardrails-service API is being replaced by North's own. Leaving externalApi.enabled: true alongside cohere-guardrails.enabled: true is unsupported: both deploy the same guardrails-service against the same AIDR instance.

Infrastructure

Gateway controller identity now lets you run multiple North installs on the same Kubernetes cluster, each with its own bundled Envoy Gateway. Set a unique controllerName and the chart keeps the GatewayClass it creates in sync automatically, and fails the install or upgrade outright if the two values ever drift apart instead of silently deploying a Gateway that doesn't match.

North API

[Alpha] Default model rules route a model to specific groups and evaluate priority top-first, via six new endpoints: list, create, get, update, delete, and reorder.

Model providers can now be managed entirely via API: eight new endpoints cover full CRUD by provider key, listing provider types and their schemas, refreshing a provider's model catalog, and testing credentials before persisting a provider.

Agents now support a draft/publish lifecycle via API, matching the versioning capability in the Admin console: PATCH /v2/agents/{agent_id}/draft to update a pending configuration, POST /v2/agents/{agent_id}/publish to make it live, and GET /v2/agents/{agent_id}/versions to list published snapshots. PATCH /v2/agents/{agent_id} is deprecated in favor of this flow.

Mobile

Increased localization coverage in the iOS mobile app now ensures a more reliable and seamless experience in all supported languages.

Compass

Compass now supports the latestParse model for image and PDF vision parsing in North.

Configurable dense embedding dimensions let you create indexes with a smaller supported vector size when the model allows it, which can reduce storage and improve performance.

Compass model usage metrics for embedding, vision parsing, and rerank calls are now included alongside other billable usage in North dashboards.

Keyword search now includes document filenames in lexical search so queries that mention words from a file's name are more likely to find it, without changing embedding-based retrieval.

Ingestion for small connector files is now faster, as Atlas webhooks send small content inline, avoiding an extra download for items like email bodies and short text attachments.

Clearer errors for password-protected files return a user-friendly reason when an upload cannot be opened because it is password-protected or otherwise unreadable.

More reliable connector downloads. Atlas download timeouts have been raised to 300 seconds, with separate read and connect limits, reducing transient failures on slow exports.

Original source

North Platform v1.14.6 Release

Version Information

Platform Version: v1.14.6 (previous: v1.14.5)

North App Version: v0.350.9 (previous: v0.350.8)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Bug fixes

Addresses failed saves on admin white-label configurations on S3-compatible stores that do not support conditional writes.

Fixes the Automations page failing to load when any automation contains a node that is no longer valid, so the list of automations still loads successfully.

Prevents install and upgrade failures caused by invalid toolkit-frontend Deployment YAML syntax when white-label and custom frontend init containers are both configured at the same time.

North Platform v1.13.13 Release

Version Information

Platform Version: v1.13.13 (previous: v1.13.12)

North App Version: v0.329.15 (previous: v0.329.14)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Ensures My files image uploads use the system-level image settings instead of a client-side allow_images flag, so images upload successfully when enabled.

Prevents install and upgrade failures caused by invalid toolkit-frontend Deployment YAML syntax when white-label and custom frontend init containers are both configured at the same time.

North Platform v1.12.26 Release

Version Information

Platform Version: v1.12.26 (previous: v1.12.25)

North App Version: v0.305.29 (previous: v0.305.29)

Compass Chart Version: v5.4.1 (previous: v5.4.1)

Compass App Version: v0.73.8 (previous: v0.73.8)

Bug fixes

Prevents install and upgrade failures caused by invalid toolkit-frontend Deployment YAML syntax when white-label and custom frontend init containers are both configured at the same time.

Original source

North Platform v1.14.5 Release

Version Information

Platform Version: v1.14.5 (previous: v1.14.4)

North App Version: v0.350.8 (previous: v0.350.7)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Bug fixes

Restores readable text in Terms of Service and banner content by overriding pasted inline colors so dark-mode text no longer renders white-on-white (or otherwise unreadable).

North Platform v1.13.12 Release

Version Information

Platform Version: v1.13.12 (previous: v1.13.11)

North App Version: v0.329.14 (previous: v0.329.13)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Restores readable text in Terms of Service and banner content by overriding pasted inline colors so dark-mode text no longer renders white-on-white (or otherwise unreadable).

Forces a fresh (non-cached) permission check when opening or running an agent, eliminating the delay that could return 403 errors for minutes after an agent was made public.

Original source

Outsized performance, right-sized footprint — translation built for speed and cost efficiency.

Today, we're releasing North Small Translate, a mixture-of-experts machine translation model with strong performance across 50+ languages. Across WMT26 benchmarks,¹ North Small Translate achieves an 83.6 score across all languages, outperforming proprietary models like DeepL and Google Translate, as well as open-weight alternatives such as Gemma 4 31B (off), GLM 5.2, and Mistral Large 3.

North Small Translate marks a significant milestone as Cohere's first translation model in the North model family. It builds on our multilingual and translation lineage — from the Tiny Aya model family to Command A Translate — and represents a clear next step in Cohere’s commitment to offering high-quality machine translation wherever it is needed.

Now available for research and non-commercial use under a CC BY-NC 4.0 license, North Small Translate advances Cohere’s mission to make sovereign AI a technological reality.

Visit Hugging Face to download the weights - available in several near-lossless quantizations - explore our HuggingFace space to demo the model, and read our implementation guides.

Snapshot

• Model: North-Small-Translate-1.0
• License: Open-weights, non-commercial
• Architecture: MoE
• Model size: 218B total; 25B active
• Context length: 16k input, 16k output
• Input modalities: Text
• Output modalities: Text
• Languages: Supports 50+ languages. Full list
• Optimized for: Machine translation
• Hardware (minimum): 1× B200 @ W4A4
  2× H100s @ W4A4

Leading translation quality in the open ecosystem

North Small Translate outperforms similarly sized open-weight models under 1T parameters and API-based translation models in various dimensions of machine translation on average. In WMT model evaluations, North Small Translate leads with an WMT26 All Languages benchmark score of 83.60, compared with 81.56 for Qwen 3.5 397B A17B, 76.50 for GLM 5.2 FP8, 81.37 for DeepL NextGen, 79.46 for Gemma 4 31B (on), and 68.20 for Google Translate. North Small Translate (Agentic) — which can find errors and fix errors in translation — scores even higher, at 84.36.²

North Small Translate performs strongly across 32 high-resource languages and 18 additional languages. North Small Translate is the most consistent performer across the full spread of regions, without the sharp regional drop-offs seen in other models of its size. On average across all languages, it is the best-performing dedicated machine translation model in this evaluation — open or closed.

At the regional level, North Small Translate punches above its weight, beating Gemma 4 31B (on) outright in Europe (82.2 vs. 73.9) while running essentially even with it in South Asia (86.2 vs. 86.7).

At the regional level, both North Small Translate and its Agentic counterpart beat Gemma 4 31B (on) outright across Europe — EU languages (82.74 Agentic / 82.17 standard vs. 72.73) and non-EU European languages (81.52 / 81.24 vs. 75.90) — while running essentially even with it in South Asia (87.13 / 86.16 vs. 88.04).

Both versions also outperform DeepL NextGen across every non-European region tested — MENA, South Asia, Southeast Asia, and East Asia — with the largest advantage in South Asia and MENA (roughly 8–10 points ahead of DeepL), a moderate edge in Southeast Asia (about 4–5 points), and the narrowest edge in East Asia (about 1-3 points, with the standard model closing in on DeepL's 85.41 score).

Increased throughput for faster workflows

North Small Translate is built for high-throughput generation, prioritizing raw output speed even as concurrency scales.

In our testing, North Small Translate achieved up to 1.4x higher output throughput than Gemma 4 31B TP1 (1 x GPU) under identical concurrency levels and hardware configurations — 112 vs. 81 Output Tokens per Second (TOPS) at low concurrency and 39 vs. 30 TOPS at high concurrency. In practical terms, that's 30-38% more tokens generated per second, translating to meaningfully faster completion times on longer outputs.

Long documents are where many translation models fall apart, and North Small Translate isn't one of them. It scores 48.9 on our long-context evaluation, more than double Google Translate (21.3) and Gemma 4 31B (19.4), and ahead of every general-purpose LLM we tested.³

Paired with its throughput advantage, that means fast, reliable translation at length, without the quality collapse seen in most non-specialized alternatives.

Efficient cost of translation, at scale

Efficiency is a core constraint in enterprise translation deployment, and we engineered North Small Translate to be extremely cost-efficient, without sacrificing performance.

For enterprises evaluating commercial licenses of North Small Translate, benefit from a strong 80.1 score at just $0.000676 per task, using only 661 tokens on average. Compared to Gemini 3.1 Pro Preview (high) — which costs $0.038928 per task (5,762% more than North Small Translate).

Similarly sized models like Qwen 3.5 397B A17B and Cohere’s own Command A+ have decent performance at $0.004525 and $0.005158 per task, respectively.

Best in translation, in partnership with RWS

North Small Translate was developed in partnership with RWS, an AI solutions company pioneering in language technology and services. Close collaboration with RWS, specifically Language Weaver’s research and science teams along with its language experts, helped shape the model's real-world translation performance throughout development. RWS works with more than 80% of the world’s top 100 brands, empowering the world's most ambitious brands to communicate seamlessly across borders and cultures.

For enterprises that need more than open-weight research access, security, scalability, and a dedicated translation and localization platform, North Small Translate, is available through RWS’s Language Weaver product.

Getting Started

North Small Translate is available today on Hugging Face for non-commercial and research use. Visit our documentation for detailed model specs, deployment guides, and implementation examples to get started.

Footnotes

¹ WMT Benchmarks focus on evaluation of general capabilities of machine translation (MT) systems. Its primary goal is to test performance across a wide range of languages, domains, genres, and modalities.

² To understand WMT Benchmark scores, the scoring methodology defines performance ranges as follows: 0-20 (not acceptable), 20-40 (borderline), 40-60 (acceptable), 60-80 (good with major errors), and 80-100 (perfect or with minor errors).

³ Long context evaluation measures how well a model can translate two chapters of a book on a single call. The quality is measured for each paragraph in isolation via xComet-XL metrics.

Original source

We're pleased to announce the release of North Small Translate, an open-weights mixture-of-experts model purpose-built for machine translation across more than 50 languages.

North Small Translate is designed to give researchers, developers, and enterprises flexible ways to evaluate and deploy machine translation while retaining control over their data and infrastructure.

Key features

• Purpose-built translation: Optimized for machine translation across more than 50 languages and locale variants.
• Efficient MoE architecture: 218 billion total parameters with 25 billion active parameters.
• Flexible deployment: Available through the free-tier Chat V2 API and as open weights in W4A16, FP8, and BF16 for non-commercial use.
• Private deployment: Suggested deployment hardware by quantization format:W4A16: Two H100s or one B200
  FP8: Four H100s or two B200s
  BF16: Eight H100s or four B200s

Technical details

• Model name: north-small-translate-1-0
• Context length: 16K
• License:
  Creative Commons Attribution-NonCommercial 4.0
• Open-weights formats: W4A16, FP8, and BF16

Availability

North Small Translate is available on the free tier through the Chat V2 API. Open weights are available in W4A16, FP8, and BF16 on Hugging Face for non-commercial use under the CC BY-NC 4.0 license.

For supported languages, use cases, and an API example, see the model documentation.

Original source

North Platform v1.14.4 Release

Version Information

Platform Version: v1.14.4 (previous: v1.14.3)

North App Version: v0.350.7 (previous: v0.350.6)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.80.0 (previous: v0.80.0)

Bug fixes

Fixes truncated user and group names in the admin permissions list by adding a hover tooltip that reveals the full name.

North Platform v1.13.11 Release

Version Information

Platform Version: v1.13.11 (previous: v1.13.10)

North App Version: v0.329.13 (previous: v0.329.12)

Compass Chart Version: v5.7.0 (previous: v5.7.0)

Compass App Version: v0.76.0 (previous: v0.76.0)

Bug fixes

Fixes truncated user and group names in the admin permissions list by adding a hover tooltip that reveals the full name.

Ensures utility LLM calls (title generation, memory extraction, AI suggestions) route through the correct model by syncing the model ID onto the request context before provider calls.

North Platform v1.12.25 Release

Version Information

Platform Version: v1.12.25 (previous: v1.12.24)

North App Version: v0.305.29 (previous: v0.305.28)

Compass Chart Version: v5.4.1 (previous: v5.4.1)

Compass App Version: v0.73.8 (previous: v0.73.8)

Bug fixes

Forces a fresh (non-cached) permission check when opening or running an agent, eliminating the delay that could return 403 errors for minutes after an agent was made public.

Original source