In March, Anthropic unveiled the models Claude 3but Europe has been left out since their introduction.
In just two months, thanks to support from Amazon and Google, the artificial intelligence company has adapted its work to comply with the strict regulations of the AI Act, which comes into force next year.
Now, the Claude AI bot is officially available in Europe, both as a web application and via the iOS App Store, with both free and paid offerings.
Its computational capacity extends up to 200,000 tokens, the equivalent of a tome of approximately 500 pages.
Three versions, three models
The chatbot, now in its third version, has made a significant leap forward.
Anthropic claimed greater accuracy, fewer errors, and a better ability to understand and answer complex, factual questions, with a negligible rate of rejection and rejection of requests.
The Claude 3 family includes three models with different performances: Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus.
The 200,000 token context referenced above, in chatbots, refers to the maximum length of the text inserted as input.
Considering this interval and an average of approximately 300 words per page, it explains why Claude AI is considered capable of “absorbing” a text approximately 500 pages long.
In comparison, ChatGPT, depending on the version used, ranges from 4 to 128 thousand tokens.
Anthropic says Claude 3 shows strong understanding and fluency in European languages such as French, German, Spanish and Italian.
In the data sheet presented in March, the company claims a score of 80% on the 5-shot Multilingual MMLU (Reasoning).
This term indicates the accuracy of the AI models in the tasks on a specific benchmark: 5 shots means that the model was given 5 complete pairs of questions and answers as demonstrative examples before evaluating its performance on the remaining test instances for the task reasoning on the multilingual MMLU dataset.
While Haiku represents the entry level, Sonnet is presented as the chatbot that offers an ideal balance between intelligence and speed, with high performance at a lower cost than competitors.
Haikuthe least advanced version, is suitable for improving interactions with corporate customers and content moderation.
The most advanced model, Opuscan even reach a 1 million token context for select customers.
According to the developers, this version demonstrates “Near-human levels of understanding and fluency in complex tasks” and outperforms the competition in most assessment benchmarks, including undergraduate-level expert knowledge (MMLU), graduate-level expert reasoning ( GPQA) and basic mathematics (GSM8K).
Costs and availability
Users have been able to access Claude via API since the beginning of the year, allowing European developers to integrate models into applications, websites and services.
Claude.ai and the Claude iOS app are available for free, but users can subscribe to Claude Pro for 18 euros per month + VAT to access all templates.
A collective plan is also available at 28 euros per user per month, with a minimum of 5 users.
Plan Claude Proat a cost of 21.96 euros per month, allows access to all models, including Claude 3 Opus.
Similarly, a Team Pro plan is available at 34.16 euros per user per month, with a minimum of 5 users.
Users with a free account for Claude AI can only use the Claude 3 Sonnet model and have fewer total inferences than the Plus plan, at least 5 times fewer.
Claude will still warn users before reaching the 7 remaining messages and will reset the limit every 5 hours once they are finished.
