After a summer of chaos, Openai fell

Openai has just spent his best week for months. And he desperately needed it.

The company based in San Francisco, better known for Chatgpt, has spent a large part of June and July in the headlines for all bad reasons. The talent raid came for the first time: the meta-PDG Mark Zuckerberg opened the checkbook, offering hundreds of millions of dollars in compensation to attract the best researchers from Openai. Several ships have jumped. The CEO, Sam Altman, was unleashed publicly, calling the mercenary of Meta’s approach and accusing him of not having culture.

Then came the failed Windsurf acquisition, a hot AI startup specializing in the infrastructure of ai-native data, which Optai had been in talks to buy. Google plunged at the last minute and concluded the agreement instead, a humiliating loss in the AI arms race with high issues.

And to top it all, Openai had to delay the release of its open source models promised for a long time after intense pressure from developers, fueling criticisms that the company took competitors like Meta, which aggressively published its own models for free.

Internally, things seemed chaotic. Leadership has given all employees a week off and disclosed memos described a besieged company, a fortress attacked on all sides, or worse, a fire house. The darling of AI, in the past, began to look shaken, and the perception that Meta had stolen its momentum increased.

From panic to pivot

This week, Openai finally started playing the offensive again. First of all, he published the long-awaited open-source models, a decision to appease developers and reaffirm his relevance in the open IA ecosystem. Three days later, the largest swing: the launch of GPT-5, presented as the most powerful chatbot on the market.

OPENAI says GPT -5 addresses two of the greatest complaints concerning AI assistants: “hallucinations” – when chatbots spit false information with confidence – and the too polite and bland tone that makes them look like public relations trainees. The company claims that the new model is faster, more precise and capable of providing more nuanced answers without sugar. By learning to say “I don’t know”, GPT-5 aims to be the first IA chatbot in which you can really trust.

Although independent tests are necessary to confirm these claims, the deployment has given Openai something he has not had for weeks: control of the story. For the moment, the AI projector is back in San Francisco, not in Menlo Park, where the so-called “dream team” of Meta of the ex-openai researchers builds its own models.

At the same time, the company is in discussion with investors on a massive action sale of employees which would value it at $ 500 billion, a decision widely considered as a defensive strategy to create “golden handcuffs” and stop the exodus of talents.

The big question: was it just a good week, or the start of a real return? In the world at high speed of AI, stability rarely lasts a long time.

While OpenAi’s ambitious statements on GPT-5 must still be checked, this week’s message was undoubtedly: while its rivals wrote checks and poaching, Openai built. With these two major launches, the company has actually regained conversation control.

The Ai tag always shines the brightest in San Francisco, not in Menlo Park, where “the dream team” of Meta des Mercenaires Ai is based. The question is now whether this powerful demonstration of force is sufficient to end the distractions and definitively find the momentum.