The gap between the top AI models has shrunk to almost nothing. MIT Tech Review published a data-driven snapshot of where the AI industry stands right now, and the picture is striking: Anthropic leads the pack as of March 2026, with xAI, Google, OpenAI, and Chinese labs like DeepSeek and Alibaba trailing by razor-thin margins on Arena, the community-driven LLM ranking platform.
Rewind to early 2023, and OpenAI had a comfortable lead with ChatGPT. That feels like ancient history. Google and Anthropic closed the gap through 2024, and DeepSeek’s R1 model briefly matched ChatGPT in February 2025. The leaderboard now reshuffles constantly.
What stands out here is the shift in what “winning” means. When every top model performs roughly the same on benchmarks, the competition moves to cost, reliability, and real-world usefulness. Raw capability is table stakes.
🇺🇸 vs 🇨🇳 Different Strengths, Same Race
The US and China are nearly tied overall, but they’re winning at different things, according to MIT Tech Review. The US holds advantages in model power, capital, and infrastructure (5,427 data centers, more than 10x any other country). China leads in AI research publications, patents, and robotics.
This split matters. The US builds the best models today, but China is laying groundwork for tomorrow’s applications. Robotics and patents translate into long-term industrial advantage. Neither side can claim dominance.
🔒 Transparency Is Disappearing
Here’s a trend that should concern everyone in the field: OpenAI, Anthropic, and Google have all stopped disclosing training code, parameter counts, and dataset sizes. The era of open research papers accompanying major model releases is fading fast.
“We don’t know a lot of things about predicting model behaviors,” says Yolanda Gil, a computer scientist at USC who coauthored the report. She warns this lack of transparency makes it harder for independent researchers to study AI safety.
This is significant because safety research depends on understanding what’s inside these systems. As models get more capable, the people building them share less about how they work. That’s a tension the industry hasn’t resolved.
📈 Still No Plateau in Sight
Despite repeated predictions that AI progress would hit a wall, it hasn’t. MIT Tech Review highlights that AI models now meet or exceed human expert performance on PhD-level science, math, and language tests. Software engineering benchmarks (SWE-bench Verified) jumped from ~60% in 2024 to nearly 100% in 2025. In 2025, an AI system independently produced a weather forecast.
“I am stunned that this technology continues to improve, and it’s just not plateauing in any way,” says Gil.
But the progress is uneven. AI still exhibits what researchers call “jagged intelligence.” Robots succeed in only 12% of household tasks. Self-driving is further along (Waymo operates in five US cities, Baidu’s Apollo Go runs in China), and AI is pushing into law and finance, but no model dominates any professional domain yet.
🎯 What This Means for Practitioners
- Model selection is now about fit, not rankings. When the top five models are nearly identical in capability, pick based on cost, latency, API reliability, and how well a model handles your specific use case.
- Bet on applications, not model providers. The value is shifting from “who has the best model” to “who builds the best product on top of these models.”
- Watch the transparency gap. If your work depends on understanding model behavior (safety, compliance, regulated industries), the shrinking disclosure from major labs is a real risk factor.
- China’s robotics lead is underrated. Most Western AI coverage focuses on LLMs. The patents and robotics research happening in China will matter more as AI moves from text to physical-world applications.
The AI race isn’t slowing down. It’s just changing shape. For the full data and charts behind these trends, check out the original analysis at MIT Tech Review.