Google DeepMind and Xiaomi MiMo are the biggest gainers in Agent Arena as tasks become more complex, while OpenAI and Anthropic continue to dominate the frontier positions on cheaper tasks.
Agent Arena · Frontier rankings by task cost
As tasks get harder, the frontier opens up beyond OpenAI and Anthropic
On lower-cost tasks, OpenAI and Anthropic hold 10 of 11 frontier spots. On the most expensive 5% of sessions, that falls to 7 of 11. Google DeepMind and Xiaomi MiMo are described as the clearest gainers.
10 of 11
OpenAI and Anthropic frontier spots on lower-cost tasks
7 of 11
OpenAI and Anthropic frontier spots on the most expensive 5% of sessions
4 of 11
Frontier spots held by other labs on the most expensive sessions, up from 1
Frontier spots by task cost (each block = 1 of 11 spots)
Blue = OpenAI and Anthropic · Green = other labs, including Google's Gemini 4
Lower-cost tasks
10 OpenAI/Anthropic · 1 other
Most expensive 5%
7 OpenAI/Anthropic · 4 other
What the pattern suggests
The best model for cheap, simple work is not necessarily the best for demanding, long-running workloads.
Gainers
Google DeepMind and Xiaomi MiMo, as complexity rises
Key newcomer
Gemini 4 has taken positions previously held by the two leading U.S. labs
Open question
Whether newer entrants can hold their spots on the most expensive sessions
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...