← Back to archive

Monday, May 25, 2026

ChatGPT just solved an open math problem

OpenAI just dropped new customizable audio models for voice agents while Google countered with managed AI agents running full Linux environments on Gemini API (wild timing). Meanwhile, ChatGPT o1 Pro apparently solved an open PhD-level math problem in two hours—mathematician Timothy Gowers is calling it the first really clear example of AI cracking a well-known unsolved problem. When AI starts solving problems that stump human mathematicians, should we be excited or terrified?

Top Stories

1
OpenAI Introduces New Audio Models in the Api

OpenAI launched new speech-to-text and text-to-speech models in its API with state-of-the-art accuracy and first-ever instruction-based voice customization, enabling developers to build more powerful and personalized voice agents for customer service, transcription, and creative applications.

openaiaudio-modelsspeech-to-texttext-to-speech
2
Timothy Gowers calls it the first really clear example of AI solving a well-known open math problem

Twitter/X

ChatGPT 5.5 Pro autonomously solved open mathematics problems at PhD-level in under two hours, with experts confirming its approach was completely original. This represents a watershed moment where AI capabilities may fundamentally reshape what constitutes meaningful mathematical research contribution.

llmopenaimathematicsresearch
3
Cohere Command A+ Open Release

Twitter/X

Cohere released Command A+ as an open-source MoE model under Apache 2.0, delivering major performance gains in agentic tasks, multimodal understanding, and multilingual capabilities while enabling efficient deployment on minimal hardware. This release advances sovereign AI by giving enterprises full control over enterprise-grade agentic capabilities.

cohereopen-sourcemoeagents
4
OlmoEarth v1.1: A more efficient family of models

Hugging Face

OlmoEarth v1.1 cuts compute costs by up to 3x compared to v1 by using a more efficient tokenization method for satellite imagery, while maintaining similar performance on remote sensing benchmarks. This enables more organizations to deploy planet-scale environmental monitoring and mapping applications.

remote-sensingtransformersefficiencyopen-source
5
Google launches managed agents on Gemini API with built-in Linux environments

Google introduced Managed Agents on Gemini API, providing developers with hosted AI agents that include Google-managed Linux environments accessible via single API calls, with customizable instructions and tools defined in Markdown.

googlegeminiagentsapi

Keep Reading

Enjoyed this issue?

Get daily AI intel delivered to your inbox. No fluff, just the stories that matter.