Canadian AI startup Cohere has raised 500 million dollars (US) from investors, or just under 460 million euros. That is against a valuation of 5.5 billion dollars (5 billion euros). With that, the company has more than doubled its value from last year.
The investors include a major asset manager and a host of companies such as Fujitsu, investment vehicle AMD Ventures of the chip company of the same name and Cisco, which also already has a partnership with Cohere to build LLMs (just as it has with Mistral AI).
Bloomberg reports that as recently as last year, Cohere’s valuation was 2.2 billion dollars, when it raised 270 million in a similar investment round. With these and earlier investments, the total amount of money invested in the company has reached 970 million dollars.
No chatbot for wide audience
Unlike competitors like Open AI, Anthropic, Mistral AI and Google, Cohere lacks imaginative chat functionality like ChatGPT, Claude, Le Chat and Gemini. Nor does the company offer a generally accessible image generator. Instead, the company relies on specific applications for business use cases, for which it builds LLMs from the ground up.
In its own words, the company is not looking for the holy grail called AGI (Artificial General Intelligence, or AI that is as good or better at tasks across the board as humans) but wants to “…make models that can be efficiently run in an enterprise to solve real problems”, said Nick Frosst, one of Cohere’s co-founders.
Founded in 2019, Cohere has experienced stormy growth in recent years. Co-founder Aidan Gomez was one of the authors of the research paper “Attention Is All You Need” that launched the current AI era. The company’s most recent model is Command R+, which primarily helps companies process and interpret large volumes of production workloads. Another LLM from the startup is the open-source Aya 23, which has 23 languages and is available in variants with 35 billion parameters or a compact 8 billion.
Tip: Cohere introduces Aya model that speaks 23 languages
Less hallucinating
Flagship Command R+ offers enhanced Retrieval Augmented Generation (RAG) functionality. This ensures, for example, that it ‘hallucinates’ less and answers become much more accurate. The LLM searches private databases for proprietary information and uses this to improve the accuracy and applicability of its answers.
Cohere said Command R+ is intended to compete with rivals such as OpenAI’s GPT while costing less. Nvidia and Oracle, among others, previously invested in the startup. Oracle has a partnership with Cohere that currently allows it to use only that company’s foundation models.
Also read: NetSuite Enterprise Performance Management (EPM) now available in EMEA