Model News Q2 25

This article summarizes notable developments and updates in AI sota foundation models during Q2 2025.

Also the second quarter of 2025 saw major tech players launching new systems and startups pushing innovation.

Major Releases

OpenAI: o3, o4, GPT-4.1

• Released o3 and o4-mini April 16, 2025. See Open AI announcement:
  • o3: OpenAI’s most advanced reasoning model, strong at coding, math, science, and visual tasks
  • o4-mini: Smaller, cost-efficient, top performer on AIME 2024/2025 math competitions
  • This are OpenAIs first modesls that think in images
  • More details: TechCrunch article
• Releases the GPT-4.1 models which outperform GPT‑4o serie.

Anthropic:

• unveiled its next-generation models Claude 4, which emphasizes advanced reasoning and expanded the AI’s context window to 200,000 tokens
• releases Claude 3.5 Sonnet

Open Models: Meta and more China models

• Meta released Llame 4
• Mistral releases its first reasoning model Mistral Magistral
• DeepSeek-V3 model released with top scores on Chatbot Arena open-source leaderboard
• Alibaba released updated Qwen models with top scores