Hey Everyone,
I’m watching Cohere’s situation really closely in 2024. I’m a fan of this team but I’m very concerned about funding to revenue generation as a ratio at this company.
Canadian pension investment manager PSP Investments is set to co-lead a funding round of at least $500 million in Toronto-based Cohere that would value the AI startup at $5 Billion, not quite x.AI ambition, but still. As of last year they were generating around $13 million in ARR according to The Information.
So Command-R is actually fairly impressive.
This week it’s trending on HuggingFace:
Sara Hooker mentions that they have over 90+ thousand downloads since it was released C4AI Command R+ last Friday.
Harning the Power of RAG
Command-R represents a significant leap forward for Cohere’s technology, offering enhanced performance on key AI tasks like retrieval augmented generation (RAG) and tool use, longer context windows up to 128,000 tokens, and more affordable pricing.
Revenue growth might have a good 2024 by the looks of it. The Toronto-based company has seen its annualized revenue run rate grow to $22 million this month from $13 million in December as it launched new model Command-R, according to Reuters.
Features
Advanced Retrieval Augmented Generation (RAG) with citation to reduce hallucinations
Multilingual coverage in 10 key languages to support global business operations
Tool Use to automate sophisticated business processes
A 128k-token context window and is designed to offer best-in-class
RAG and Adaptive RAG are sort of a big deal.
Cohere have now partnered with Microsoft Azure, which should help their monetization issues a bit.
⌘-R
Introducing Command-R, a model focused on scalability, RAG, and Tool Use. We've also released the weights for research use, we hope they're useful to the community!
Highlights
Command R boasts:
Strong accuracy on RAG and Tool Use
Low latency, and high throughput
Longer 128k context and lower pricing
Strong capabilities across 10 key languages
Model weights available on HuggingFace for research and evaluation
Command R will be available immediately on Cohere’s hosted API, and on major cloud providers in the near future.
Leaders in RAG
Industry Leading RAG Solution
It’s a good fit for Enterprise customers, here’s why:
RAG has become a foundational building block for enterprises adopting LLMs and customizing them with their own proprietary data. Command R+ builds upon Command R’s exceptional performance at RAG use cases.
Retrieval-augmented generation (RAG) has become a crucial pattern in the deployment of LLMs. RAG enables enterprises to give the model access to private knowledge that it otherwise would not have.
By letting the model search over private databases and use that information to form responses, the accuracy and usefulness of the model changes dramatically. The key components to RAG are:
Retrieval: Searching over corpora of information relevant to responding to a user.
Augmented Generation: Using the information retrieved to form a more informed response.
⌘R+ looks like it's the fastest growing model on Open Router. See “Fastest Growing” here.
Tool Use is now available via their API, read more here.
O’ Canada 🍁
Canada recently announced $2.4 Billion in Federal AI funding for AI startups, compute and saftey considerations for 2024. Cohere certainly does deserve some of that because in some ways, it’s the closest Canada has to their own OpenAI.
Securing Canada’s AI advantage
According to the CBC, the bulk of that — $2 billion — is going into a fund that will aim to provide access to computing capabilities and technical infrastructure.
Multilingual Support for Global Business Operations
You can feel this with Cohere’s ethics and concern for issues.
Command R+ is designed to serve as many people, organizations, and markets as possible. During their discussions with companies, they met with a huge demand for multilingual capabilities that helps organizations more seamlessly work across regions and cultures. Some are calling this the push for Sovereign AI, to protect cultural, national, linguistic, religious and distinct heritage.
Being valued at 200 times its forward revenue is pretty incredible for Cohere. They must have one hell of a product timeline to get away with that - the Canadian Pension Fund, just wow okay. i.e. (Canadian Pension Fund PSP).
Still I can’t help myself but want to cheer for Cohere and Command-R looks seriously impressive.
Some of their top tier customers include Salesforce, Oracle, Jasper and Notion. The round values Cohere at around $5 billion, but the deal isn’t finalized and that number could change.
Democratizing RAG AI
Retrieval augmented generation (RAG) is a natural language processing (NLP) technique that combines the strengths of both retrieval- and generative-based artificial intelligence (AI) models.
On Azure, accessing Command-R+ costs $0.015/$0.003 per 1000 tokens out/in (the price for getting the model to respond with its own outputs and for entering in prompts, respectively.
Not only is Command R+ a strong multilingual model, but the R-series of models features a tokenizer that compresses non-English text much better than the tokenizer used for other models in the market, capable of achieving up to a 57% reduction in cost.
On Amazon Sagemaker, it is billed differently at $82.89 to $146.93 per hour for running it in real time inference (using the model in production for end users).
Microsoft’s blog about Cohere Command-R was pretty bullish. Aiden Gomez and Sara Hooker in my mind are some of the most likeable people in Generative AI.
Cohere is as humble as OpenAI is full of bravado, true to their countries of origin. But don’t underestimate the technical chops of this team, many worked at Google before. Aiden worked with Geoffrey Hinton and Łukasz Kaiser on the Google Brain team.
The worst thing about this startup is the font they use on their blogs.
I consider them leaders in:
RAG implementation
Data Privacy and Security
Cost
LLMs designed for use at production scale
For Cohere, getting placement on more leading cloud and AI development platforms hopefully means they are on the fast track of Enterprise AI deals and increased revenue. It’s a hyper competitive market now and they need to do have a big 2024 and 2025 in order to survive.
Endorsements
LangChain
Scale AI
Command-R is available now on their playground and it will be up on all major clouds shortly: https://dashboard.cohere.com/playground/chat