While Chatgpt collapses with GPT-5, Anthropic chooses this moment to wear the blow of grace with Claude 4.5

The disappointments accumulate around GPT-5. Too expensive, too unstable, too far from the real needs of users … Critics accumulate, and it is in this context that Anthropic launches Claude 4.5. This update does not promise to revolutionize Artificial intelligencebut simply to better meet the concrete expectations of professionals.

Claude 4.5 is there: what innovations?

If Anthropic Announcement “The best coding model in the world”Claude 4.5 focuses on the essentials: reducing code errors, more finely integrated into the versioning and debugging tools of developers, and more naturally enters into collaborative work platforms. The internal anthropic benchmarks show increased precision, especially in programming, where the model now displays a drastically reduced error rate on certain targeted tasks. Claude Pro’s price remains set at € 21.60 per month and Claude Max, from € 108.

However, despite this progress, the model remains limited in volume. Even for paid subscribers, the context window caps at 200,000 tokens (around 150,000 words). It is a constraint that can slow down the most ambitious projects, especially in the face of competitors like Gemini 2.5 Pro, which offers up to 2 million tokens. Anthropic also maintains his restrictions: filtering of queries, refusal to deal with certain sensitive subjects, and a policy of moderation which can frustrate those who seek a freer AI.

GPT-5: Air promises and disappointed users

But here it is: what strikes this outing is its timing. Because during this time, Openai launched GPT-5 with a lot of noisebut Returns are far from up. Despite Sam Altman’s announcements on a model “Doctoral level”users note persistent basic errors, inconsistent reasoning, and practical limitations that make the tool unreliable on a daily basis.

Without evoking certain unexpected restrictions, such as automatic re-retaliations to the new model for subscribers. Result: a lot talk about regression compared to GPT-4Oand the forums are full of harsh criticism.

Faced with these disappointments, some explore other solutions. Mistral offers open and more flexible models, often cheaper, which attracts those who want an AI without too much restrictions. Google, with Gemini 2.5 Pro (€ 21.99 per month), focus on a massive context window, ideal for projects requiring processing large volumes of data. Finally, Mistral, via its Le Chat interface, offers a less formatted approach, with a more natural tone, and a price set at € 17.99.