GPT Protocolの‌価格GPT

OpenAI officially announced that GPT-4 will be retired at the end of this month, to be completely replaced by 4o.

Meta has just unveiled Llama 4, its most advanced artificial intelligence model to date. According to David Sacks, AI advisor at the White House, this technological advancement restores the United States’ advantage in the global competition for AI supremacy. David Sacks, special advisor for AI and crypto to the Trump administration, praised Meta’s advancements regarding their new AI model. In a message posted yesterday on his X account, he stated that “Meta’s Llama 4 puts the United States back in a dominant position in the race for artificial intelligence”. “For the United States to win the AI race, we must also excel in open source, and Llama 4 puts us back in the lead”, he asserted , in a context where technological rivalry is intensifying between Washington and Beijing. This statement comes shortly after his appointment following President Trump’s inauguration in January 2025. Sacks had already emphasized the importance for the United States to remain vigilant despite its lead in this strategic field. The launch of Llama 4 is part of a global innovation dynamic where major powers are investing heavily in AI, now seen as a matter of national sovereignty, comparable to energy or military sectors. Meta asserts that its new Llama 4 Scout and Llama 4 Maverick models are “the most advanced to date” and “the best in their class for multimodality.” These claims are supported by performance tests that position Meta’s models ahead of their main competitors. The Llama 4 Scout model, equipped with 17 billion active parameters and utilizing 16 experts, would outperform competitors like Gemma 3, Gemini 2.0 Flash-lite, and Mistral 3.1 “across a wide range of recognized benchmarks“, according to data shared by the company. Even more impressive, Llama 4 Maverick, also equipped with 17 billion parameters but configured with 128 experts, would match or exceed the performance of GPT-4o and Gemini 2.0 Flash models. Meta highlights that Maverick offers comparable capabilities to DeepSeek v3 for reasoning and coding tasks while utilizing only half the active parameters of its competitor. This breakthrough realizes Mark Zuckerberg’s ambitions, who announced in July 2024 that the Llama models would become “the most advanced in the industry” by 2025. A bet that seems well on its way to being realized, just two years after the launch of the first version of Llama in February 2023. The strengthened position of the United States in the AI race comes as other major initiatives multiply in the sector, such as the recent record fundraising of 40 billion dollars by OpenAI, bringing its valuation to 300 billion. In this context of global competition, Europe is also seeking to position itself, notably through strategic investments like the AI campus project funded by the United Arab Emirates in France .

Meta Launches Llama 4: Powerful Multimodal AI with 10M Token Context Window On April 5, 2025, Meta unveiled Llama 4, its latest open-source AI series, introducing two standout models: Llama 4 Scout and Llama 4 Maverick. Scout, with 17 billion parameters and a massive 10 million token context window, outperforms models like Gemma 3 and Mistral 3.1. Maverick is built for multimodal tasks and surpasses GPT-4o and Gemini 2.0 Flash, scoring 1417 ELO on the LMArena benchmark. Both models leverage a mixture-of-experts architecture and iRoPE technology to boost performance and efficiency. Meta is also training a larger model, Llama 4 Behemoth, which already outperforms GPT-4.5 on STEM benchmarks—even before completion. The models are available on Hugging Face, though their usage is currently restricted in the EU due to strict privacy regulations. #Meta #Llama4 #AI #MultimodalAI #GPT4 #Gemini #HuggingFace #TechNews #CoinPhoton

OpenAI founder Sam Altman announced plans to release O3 and O4-Mini before launching GPT-5. These releases are expected in the coming weeks, with GPT-5 following in a few months. Altman highlighted that the integration of all components is more challenging than anticipated, and emphasized the importance of ensuring sufficient capacity to meet the expected unprecedented demand. The improvements in GPT-5 are anticipated to exceed initial expectations.

Artificial intelligence is entering a new era. Sentient, a startup funded by Peter Thiel’s Founders Fund, launches Open Deep Search (ODS), an open-source AI search system that significantly surpasses the performance of market leaders like OpenAI’s GPT-4o and Perplexity. Sentient, a startup backed by Peter Thiel’s Founders Fund, has just crossed a decisive milestone in the artificial intelligence ecosystem. On April 2, 2025, the company unveiled its Open Deep Search (ODS) framework, a fully open-source AI search system. This innovation did not go unnoticed, as it immediately demonstrated superior performance compared to industry giants. Tests conducted on the “Frames” benchmark are particularly revealing of this advancement. This benchmark, which evaluates factuality, information retrieval, and reasoning of AI models, placed ODS at the top with an impressive accuracy of 75.3%. In comparison, ChatGPT-4o Search Preview by OpenAI only reached 50.5%, while Perplexity Sonar Reasoning Pro peaked at 44.4%. To ensure the integrity of these results, Sentient took care to isolate its researchers from the Frames test sets throughout the evaluation process. But as Himanshu Tyagi, co-founder of Sentient and professor at the Indian Institute of Science, emphasizes: An independent verification is only necessary for closed-source solutions. With our open-source approach, anyone can run our code, reproduce our results, and verify their accuracy. The democratization of AI represents the core philosophy of Sentient, but the company has also tackled one of the major challenges in the industry: how to effectively monetize an open-source model without compromising its accessibility. Sentient offers a unique solution with its fingerprint technology, allowing developers to protect their intellectual property while maintaining the fundamental openness of the model. This ingenious system directly addresses the dilemma that has previously hindered the widespread adoption of open-source solutions in the AI domain. “AI should belong to the community, not be controlled by closed-source companies,” Tyagi asserts. “We build, monetize, and provide open-source AI with a key principle in mind: singularity in intelligence but plurality in use cases.“ This strategic positioning has already sparked considerable interest, with over 1.8 million sign-ups on the waiting list even before the official launch of the platform. Last February, Sentient also conducted one of the largest NFT creation campaigns, enabling over 650,000 participants to acquire fractional ownership in its AI models. Sewoong Oh, principal researcher at Sentient and professor at the University of Washington, perfectly summarizes the company’s ambition: With a suitable architecture, open-source models can easily surpass closed software giants. The results of these benchmarks validate our mission: to create an open ecosystem that benefits all AI developers and users. The emergence of Sentient and its ODS framework aligns with a broader trend of challenging dominant proprietary models. This advancement potentially marks a decisive turning point for the industry, where open-source solutions are starting to demonstrate their ability to compete with, if not surpass, established closed platforms like OpenAI or Perplexity. This breakthrough is part of a radical transformation of the AI landscape. As OpenAI attracts $40 billion in investment from SoftBank and reaches a record valuation of $300 billion, open-source solutions like Sentient and DeepSeek R1, a Chinese model with remarkable performance, are shaking up the status quo with fractional usage costs. In summary, this confrontation between overfunded proprietary models and high-performing open-source alternatives is redrawing the economic contours of the sector. The simultaneous rise of these opposing approaches suggests a new era where accessibility and technical performance may take precedence over centralized business models, thus challenging the dominance of American AI giants.