Google this week began previewing a major new version of its large-scale language model Gemini for early testing

This is because Google claims that this new version, Google Gemini 15, is more efficient than previous versions and can "understand long contexts" with a potential context window of up to 1 million tokens

This is a technical way of measuring the amount the model can understand before reaching its limits, and it is significant that Google claims that Gemini 15 can handle up to 1 million tokens

Competitors such as ChatGPT Plus and Microsoft Copilot, which currently run on OpenAI's GPT-4 Turbo LLM, have context windows of up to 128,000 tokens Therefore, Google's claim that Gemini 15 can understand up to 1 million tokens is an important upgrade

According to Google, Gemini 15 with a context window of 1 million tokens can read, digest, and answer questions about up to an hour of video, 11 hours of audio, over 700,000 words (Tolstoy's War and Peace), or over 30,000 lines of code to understand In short, it's a big upgrade

It is important to note, however, that not everyone using Google Gemini 15 will have access to the full 1 million token context window now or at the time of release

According to a Google blog post published today (February 15), the company is offering limited developers and business customers early access to Gemini 15 and the 1 million token window (through AI Studio and Vertex AI development platforms)

However, according to Google, the rest of us will have to wait a little longer for the official release of Gemini 15 Pro And even if Gemini 15 Pro is released, the entry level will be limited to the market standard of 128,000 tokens Google has stated that it plans to "introduce a price tier" that will raise the token limit to 1M If you want to play with an AI that can watch and understand an entire Buster Keaton movie (see clip above), be prepared to pay a lot of money