97% on HumanEval coding benchmark, 95% on MMLU, and 91% on MATH-500?

97% on HumanEval coding benchmark, 95% on MMLU, and 91% on MATH-500

1M-token context window with 3x faster inference and 40% lower cost than GPT-4 Turbo?

1M-token context window with 3x faster inference and 40% lower cost than GPT-4 Turbo

Mixture-of-experts architecture with 2.2 trillion active parameters and native tool-calling for agents

Models & Releases

Artificialintelligence-newstoday June 25, 2026

⚡Google's latest model runs 3x faster and costs 40% less per token

Deep Dive

This week’s top AI news from AI News Hub covers the latest breakthroughs, trends, and expert insights from January 11 to 17, 2026.

Key Points

97% on HumanEval coding benchmark, 95% on MMLU, and 91% on MATH-500
1M-token context window with 3x faster inference and 40% lower cost than GPT-4 Turbo
Mixture-of-experts architecture with 2.2 trillion active parameters and native tool-calling for agents

Enterprise teams can now deploy faster, cheaper, and more capable AI for coding, analysis, and agent workflows.