• Google circumvents issues around which AI is best.
• Starts to develop drop-in replacement for OpenAI’s API.

Companies that launched their LLM offerings early have a significant advantage over those coming late to the party. Apart from the side-effect of having less human-created content to feed into their models, latecomers also find that early-adopters have already committed to a competitor’s platform.

Google’s Gemini may be right up there in terms of capability with household-name ChatGPT, but it simply doesn’t have the benefit of being the platform – ChatGPT – that everyone talked about exclusively for what seemed like months after its launch.

And once the early-adopters, research groups, and software developers the world over were committed to OpenAI’s offerings, the overheads of re-engineering tools and changing working methods will have kept many on Sam Altman’s platform, despite other LLMs being as good, sometimes better than and at least different from ChatGPT.

Under the hood, many pieces of software use API access to LLMs like OpenAI, leveraging a platform’s capability without end-users being aware of which exactly LLM they’re interacting with. While addressing an API can be relatively simple, changing code so that it interfaces with a different LLM can be regarded as a move that’s relatively pointless: one large language model is pretty much as good as another at present, so why rock the boat? Which AI is best is an irrelevant question once customers have made a large-enough dollar commitment.

Which AI is best is irrelevant

But large providers of cloud services like Google, Microsoft and Amazon would, naturally enough, prefer it if organisations put all their eggs into one cloud basket, creating complementary systems that are simple to deploy in concord with others under the same roof. Cross-platform interoperability and data migration from one cloud to another are deliberately not made simple – or at least, as simple as they are to start using.

Google has taken a step to tempt users away from the grandaddy LLM, OpenAI’s ChatGPT, to Gemini, by launching a new endpoint for its Gemini API that allows users to switch. While it’s still in a testing, beta phase, and does not yet completely replicate all the features of the OpenAI API, is thought it will become a drop-in replacement for organisations that want to make the move.

Google hopes that companies and organisations that have made significant investment in their GCP services will want to bring their AI compute into the Google stable, and providing a byte-for-byte interoperability at API level will mean there are fewer technical reasons not to do so. That way, Google wins several times over: it gains a customer from a competitor, the customer’s data gets funnelled into Google’s portfolio of data-for-sale, and the customer’s commitment to GCP is deepened. Customer data will also add to the training data available to Google for future iterations of Gemini.

The new API endpoint can already replace code that makes direct REST calls to OpenAI, or those made via OpenAI’s SDK (software development kit). The necessary changes at the user’s end are relatively trivial – supplying a Gemini API key and references to the chosen Gemini model instead of OpenAI’s, for instance.

At present, the only two OpenAI capabilities that Google has replicated are the Chat Completions API and the Embeddings API, and it’s not clear when feature parity between the Gemini and OpenAI APIs will be reached. However, it’s clearly on the horizon.

Both companies and their major competitors eschew the notion of a general, all-purpose API to address any LLM, as that would be seen as encouraging customers to switch from one provider to another on such whimsies as price, capability, or suitability.

Google’s move in reproducing some of its main competitor’s API capabilities is not the first step in the direction of seamless transition and customer choice. If it were, we’d expect to see much fanfare about open collaboration, something that OpenAI’s CEO, Sam Altman has made clear won’t be happening, despite the anomalous use of the word ‘open’ in his company’s name. Likewise, Google’s move is a good, old-fashioned attempt at customer poaching.