OpenAI’s acquired advantage with its language models has allowed them to significantly advance the capabilities of tools like ChatGPT. GPT-4 and Turbo have completely dominated the generative artificial intelligence landscape, and all in the absence of knowing more about the future GPT-5. However, there are other companies that are already capable of hot on their heels. One of them is, without a doubt, Anthropic with its model Claude.
The firm, created by former members of OpenAI, has just announced Claude 3 and the benchmarks are already they leave it above GPT-4, something that seemed impossible. Furthermore, like Gemini, its model also comes in three sizes: Opus, Sonnet and Haiku. Below these lines we tell you all the details.
Benchmarks point to higher performance than GPT-4
So far, GPT-4 has been the most powerful language model for several consecutive months, succeeded only by GPT-4 Turbo, a language model also developed by OpenAI. Instead, during all this time, Anthropic has made very notable advances in Claude. Let us remember that a few months ago they launched Claude 2.1 with up to 200,000 context tokens.
Image: Anthropic
Now, with Claude 3, things go further. And according to several benchmarks, the language model It is already above GPT-4 in its most advanced version. Since, as with Gemini, it also comes in three versions. Ordered from greatest to least capacity they are: Opus, Sonnet and Haiku.
Of these three versions, it will be Sonnet that we can try completely free through the Anthropic website. According to the company, Opus, the most advanced version, can reach a “close to human” level of understanding. To demonstrate it, they have published a video showing its great capabilities.
Claude 3 Opus is also superior in vision capabilities, that is, in interpreting, processing and displaying information from images. The benchmarks already put it above even GPT-4 Vision.
While Claude 3’s context window is still 200,000 tokens, like Claude 2.1, it appears to be made better use of in this update. An example of this is the precision test that Anthropic has published, with a much higher percentage in precision than previous versions of their models. With this context window we are talking about how Claude can analyze and process information from a document of almost 470 pages.
According to the Anthropic blog, this window could even reach a million tokensalthough this would mean a notable loss in the precision of its capabilities.
As the company has stated on its blog, both Sonnet and Opus are now available for use from Claude.ai and through the Claude API. The model has reached 159 countries, including Spain. Haiku, for its part, does not yet have an arrival date.
In Genbeta | This AI from Alibaba brings any photo to life by transforming it into a realistic video of someone talking or singing: that’s EMO