AI Breakthroughs: OpenAI, Meta & Anthropic’s Future for AI

OpenAI: ChatGPT-5, superintelligence, AI agents for workforce

The company that brought us ChatGPT has now set its sights on something grander: building machines that don’t just chat, but actually think and act independently.

OpenAI’s landmark moment came in August 2025 with GPT-5’s release, which stands out from previous generations. 

The company describes GPT-5 as “a significant leap in intelligence over all our previous models, featuring state-of-the-art performance across coding, math, writing, health, visual perception and more.”

Unlike its predecessors, GPT-5 operates as a unified system that can switch seamlessly between quick responses and deep reasoning, deciding in real-time whether to fire back instantly or take time to work through complex problems step-by-step.

So far, GPT-5 sets new records across mathematics (94.6% on AIME 2025), coding (74.9% on SWE-bench Verified) and multimodal understanding (84.2% on MMMU) while majorly reducing the hallucinations that plagued earlier models.

But what’s really intriguing is that OpenAI believes 2025 may see the first AI agents “join the workforce” and materially change company output.

The company’s not talking about simple automation anymore – as these systems can tackle complex, multi-step tasks from start to finish, powered by the advanced reasoning capabilities first introduced in GPT-5.

ChatGPT now reaches 700 million weekly active users, up from 500 million in March, while OpenAI hit its first US$1bn revenue month in July 2025, doubling from US$500m monthly at the start of the year.

These numbers show a technology that’s moved from novelty to necessity for millions of people.

What comes next might be even more ambitious. 

The company is beginning to turn its aim beyond traditional AGI to superintelligence in the true sense of the word – suggesting that we’re witnessing the early stages of an AI breakthrough that is designed to impact society itself.

Targets

• AI agents joining the workforce in 2025 to materially change company output
• Building superintelligence beyond traditional AGI capabilities
• Enabling unlimited intellectual capacity equivalent to everyone in 2025 by 2035

Focus ahead
Deploying AI agents across enterprise workflows and business operations.

Developing superintelligence capabilities that surpass human-level intelligence.

Expanding real-time multimodal AI interactions across voice, text and vision.

Scaling autonomous coding capabilities through enhanced Codex systems.

Building comprehensive AI safety and alignment research frameworks.

Pursuing profitability by 2029 with projected revenue targets exceeding US$100bn.

Progress
Revenue increased 3,628x since 2020, from US$3.5m to over US$12bn annually.

ChatGPT reached 100 million users within just two months of launch, making it the fastest-growing app in history.

Platform processes over 3bn daily user messages across all ChatGPT products as of August 2025.

Company employs approximately 500 staff while managing infrastructure serving 700 million weekly users.

O3 model scored 87.5% on ARC benchmark under high compute conditions, significantly outperforming previous AI systems at 53%.

Meta: Llama 4, open-source AI superintelligence, self-improving systems

Meta is now charting a different course in the AI race. 

While competitors focus on closed systems and workplace automation, Mark Zuckerberg has doubled down on open-source AI and what he calls “personal superintelligence” for everyone.

The company’s boldest move came in April 2025 with the launch of Llama 4 Scout and Maverick, the first open-weight natively multimodal models with unprecedented context support and built using a mixture-of-experts architecture.

These are a shift towards AI systems that can handle text, images and complex reasoning tasks seamlessly.

But the CEO’s ambitions stretch far beyond current models.

“Over the last few months we have begun to see glimpses of our AI systems improving themselves,” he claims, describing this as “the first step towards achieving artificial superintelligence.” 

It’s a striking assertion that suggests Meta might be closer to breakthrough AI capabilities than many realise.

The company has backed these claims with serious investment. 

Mark Zuckerberg plans to invest “hundreds of billions of dollars” into AI compute infrastructure, including multiple data centre superclusters designed specifically for training advanced AI models. 

Meta Superintelligence Labs, announced in June 2025 and led by Scale AI’s former CEO Alexandr Wang, is the company’s most ambitious AI initiative yet.

Unlike rivals pursuing centralised AI systems, Meta’s vision focuses on “building personal superintelligence for everyone” rather than directing AI towards automating work.

Targets

• Building personal superintelligence for everyone as an individual empowerment tool
• Making AI Studio the world’s leading destination for AI character creation in 2025
• Advancing speech models to become more natural, conversational and helpful

Focus ahead
Developing self-improving AI systems towards artificial superintelligence.

Expanding Llama 4 multimodal capabilities across Scout and Maverick models.

Building multiple gigawatt-scale AI superclusters including Prometheus and Hyperion.

Integrating AI into Ray-Ban Meta glasses and AR wearables.

Growing AI Studio platform for character and agent creation.

Advancing Meta Movie Gen for AI video generation and editing.

Progress
Over 85,000 Llama derivatives published on Hugging Face, a 5x increase from the start of the year.

Llama 4 models outperform GPT-4.5, Claude Sonnet 3.7 and Gemini 2.0 Pro on STEM-focused benchmarks.

Over US$1.5m awarded in Llama Impact Grants to support transformative AI applications.

Nearly 4bn monthly active users across Meta’s Family of Apps, representing nearly half the world’s population.

First supercluster Prometheus scheduled to come online in 2026, with Hyperion scaling to 5 gigawatts over several years.

Anthropic: Claude 4, AI safety, enterprise automation

While competitors chase consumer markets and demonstrations, Anthropic has quietly built something different: an AI company focused on safety, enterprise reliability and what many consider the future of AI development.

The company’s approach crystallised with the launch of Claude 4 in May 2025. 

Claude Opus 4 is positioned as the “best coding model in the world” and can autonomously work for nearly seven hours – a full corporate workday.

But Anthropic’s models cover a shift towards AI systems that prioritise constitutional AI principles and interpretable behaviour.

Both Claude 4 models can use tools like web search during extended thinking, alternating between reasoning and tool use to improve responses.

More intriguingly, when given access to local files, Claude Opus 4 becomes skilled at creating and maintaining ‘memory files’ to store key information, unlocking better long-term task awareness.

Behind this technical progress lies CEO Dario Amodei’s vision of AI’s trajectory. 

He predicts AI could eliminate half of entry-level white-collar jobs and boost unemployment to as much as 20% within one to five years, while maintaining there’s a 25% chance AI development goes “really badly.”

Yet Anthropic’s business model suggests confidence in a different path forward. 

The company hit US$5bn in annual recurring revenue by July 2025, up from US$1bn at the end of 2024, driven primarily by enterprise API usage rather than consumer subscriptions. 

With over 300,000 business customers and large accounts growing nearly sevenfold in the past year, Anthropic demonstrates that safety-focused AI can be commercially compelling.

Targets

• Ensuring democratic societies lead in AI development while preventing authoritarian dominance

• Measuring AI risks and enacting policies for economic benefit sharing from powerful AI
• Understanding how AI models operate through interpretability research breakthroughs

Focus ahead
Advancing interpretability research to understand how AI models operate internally.

Expanding Claude Code’s autonomous software development capabilities.

Building constitutional AI frameworks for safer, more reliable AI behaviour.

Growing enterprise API business and Microsoft Office 365 Copilot integration.

Developing AI safety standards and measurement tools for industry adoption.

Scaling international operations while maintaining safety-first development approach.

Progress
Anthropic called “fastest growing software company in history at the scale that it’s at” with ARR growing from US$1.4bn in March to nearly US$4.5bn in July 2025.

Workforce expanded 471% from 192 employees in 2022 to 1,097 in 2025.

70-75% of revenue comes from enterprise API usage rather than consumer subscriptions.

36% of Claude usage focuses on coding tasks, with 77% of enterprise activity targeting automation.

Claude Sonnet 4 now supports 1 million token context windows—handling 750,000 words or 75,000 lines of code in single prompts.