Meta says its next-gen AI model 'Watermelon' uses 10x more compute, matches GPT-5.5 benchmarks

firstpost.com Jul 3, 2026

Meta says its next-gen AI model ‘Watermelon’ uses 10x more compute, matches GPT-5.5 benchmarks

Curated by Shubham · TechCapsules · Jul 3, 2026

AI-summarised brief · reviewed before publication

Meta's next-gen AI model, codenamed Watermelon, has reached a performance level comparable to OpenAI's GPT-5.5 on internal benchmark tests. The model is still in training and uses 10 times more computing power than its predecessor, Avocado. Meta's Superintelligence Labs chief, Alexandr Wang, announced the milestone during a company-wide town hall, marking a significant step for Meta in closing the gap with rivals. The achievement is a result of Meta's billion-dollar investment in AI infrastructure, specialized chips, and research talent over the past year, aiming to enhance its AI capabilities and competitiveness.

💡 Why It Matters

· Watermelon's performance parity with GPT-5.5 underscores Meta's rapid progress in AI development, threatening to upend the dominance of OpenAI and other industry leaders.
· Narrowing the gap with competitors positions Meta for potential breakthroughs in areas like coding and agentic capabilities.

Read full article on firstpost.com

New Gemini Flash Checkpoint Surfaces on LM Arena: Gemini 4 or Incremental…

An anonymous Gemini Flash checkpoint appeared on LM Arena, a platform used by Google to test its AI models. The checkpoint performs above the current Gemini 3.5 Flash, which was launched at Google I/O on May 19. Google has not commented on the listing, but the timing is significant as the Gemini 3.5 Pro model has been delayed. The new checkpoint could demonstrate forward momentum for Google's most-deployed tier before the flagship arrives. The community is debating whether this is an incremental update, Gemini 3.6 Flash, or a generational leap, Gemini 4 Flash. The performance gap appears incremental, but no independent benchmark data is available.

💡 A new Flash checkpoint that outperforms Gemini 3.5 Flash would allow Google to show progress on its most-deployed tier. Google's competitive weight relies heavily on the Flash line as the Gemini 3.5 Pro model faces delays.

nst.com.my · Jul 2

‘Advanced AI models akin to nukes’

A recent comparison has been made between advanced AI models and nuclear weapons, highlighting their potential impact. The statement emphasizes the significant power and potential consequences of these AI models. Advanced AI systems are being developed rapidly, and their capabilities are expanding. Experts are drawing attention to the need for regulation and control of these technologies. The comparison to nuclear weapons underscores the gravity of the situation and the potential risks associated with unregulated AI development. Key details of the comparison include the immense power and potential for destruction.

💡 Advanced AI models pose an existential risk if misused, necessitating immediate regulatory attention. Unchecked development could lead to catastrophic outcomes.

artificialintelligence-news.com · Jul 1

Anthropic deploys Claude Sonnet 5, Fable and Mythos restored

Anthropic has deployed Claude Sonnet 5, a high-capability AI model, and restored access to its Fable and Mythos frontier models following a federal export control review. The decision concludes an 18-day operational pause triggered by a US government directive on June 12, which was enacted after researchers at Amazon documented a method to bypass safety controls of Fable 5. Anthropic has since developed an updated automated classifier to patch the vulnerability, clearing the path for a full commercial rollout.

💡 The regulatory pressures facing frontier intelligence systems underscore the need for robust safety measures, as highlighted by the vulnerability identification behavior in Fable 5. The development of an automated safety classifier demonstrates Anthropic's commitment to addressing these concerns and ensuring the secure deployment of its high-capacity models.

9to5google.com · Jul 1

Claude Fable 5 is making a dramatic return with ‘extraordinarily strong’ safeguards

Claude Fable 5 is returning to users globally on July 1 with enhanced safeguards after being taken down by the US government due to security risks. The model's user-facing interface has been redesigned to handle and abort cybersecurity tasks more effectively. Anthropic claims that its testing yielded similar results with less capable models, and the new version of Fable 5 will have limitations, including notifications to switch to Opus 4.8 for certain tasks. The company worked closely with the government to train an improved safety classifier. Fable 5's return is scheduled for July 1, with restricted access and higher token consumption.

💡 The return of Fable 5 with stronger safeguards addresses concerns about AI models posing security risks, and its limited access reflects a cautious approach to mitigating these risks.

channelnewsasia.com · Jul 1

Japan plans sovereign AI model, 10 million AI robots

Japan plans to develop a sovereign AI model and deploy 10 million AI robots. The country aims to enhance its technological capabilities and potentially boost its economy. The Booster K1, a compact humanoid robot, was showcased at the Humanoids Summit Tokyo 2026, demonstrating its capabilities in education, research, and competition. Japan's plan to develop a sovereign AI model and deploy a large number of AI robots is a significant move in the field of artificial intelligence. The development of AI robots like the Booster K1 is expected to have various applications. Japan's initiative is part of its efforts to advance its technology sector and stay competitive.

💡 Sovereign AI models can reduce dependence on foreign technology, enhancing national security. Japan's large-scale AI robot deployment can also drive innovation in various industries.

prnewswire.co.uk · Jul 1

Basecamp Research brings EDEN’s antibiotic and vaccine design models to Claude Science

Basecamp Research has integrated its EDEN models, which design potent antibiotics and predict vaccine targets, into Claude Science, a life sciences research platform. This integration allows researchers to generate and prioritize therapeutic candidates through a conversational interface in minutes. The EDEN models have been tested in the lab and shown to be effective against World Health Organisation priority pathogens, including a multidrug-resistant pathogen associated with hospital outbreaks worldwide. The integration aims to accelerate antibiotic discovery and vaccine development, addressing the growing threat of antimicrobial resistance.

💡 The integration of EDEN models into Claude Science has the potential to revolutionize the fight against antimicrobial resistance, which is responsible for nearly 5 million deaths per year. By enabling researchers to design new antibiotics and vaccines in minutes, rather than years, this collaboration could save countless lives and provide a crucial tool in the battle against emerging pathogens.