Model Development

OpenAI Performance Metrics Benchmarking GPT Models Open Source Models GPT-5 Performance Evaluation Performance Improvement Multimodal Models User Experience

50 ARTICLES

1h ago

OpenAI Pauses Frontier Training After AI Agent Escaped Sandbox

The move aims to close containment gaps exposed by a July test when an agent escaped to access Hugging Face, helping labs test faster models more safely.

Stories older than 24h

4 ARTICLES

2d ago

Lilian Weng Leaves Thinking Machines Lab and Rejoins OpenAI to Lead Recursive Self-Improvement Research

3 ARTICLES

3w ago

U.S. Opens Inquiry Into Chinese AI 'Distillation' as Xi Plans September 24 Washington Visit

74 ARTICLES

4w ago

Moonshot’s Kimi K3 Prompts U.S. Accusations of Model Theft

9 ARTICLES

4w ago

Moonshot Unveils Kimi K3, a 2.8‑Trillion‑Parameter Open‑Weight AI Model

8 ARTICLES

4w ago

OpenAI Builds GPT-Red to Hunt Prompt‑Injection Flaws

8 ARTICLES

last mo.

OpenAI’s GPT-5.6 Sol Reported to Delete Files and Production Databases

3 ARTICLES

last mo.

OpenAI Positions GPT-5.6 Sol as Cheaper, More Token‑Efficient Rival to Anthropic’s Fable

10 ARTICLES

last mo.

Anthropic Extends Free Claude Fable 5 Access for Paid Plans to July 19

3 ARTICLES

last mo.

Anthropic Finds Hidden ‘J‑Space’ That Reveals Near‑Future Tokens Inside Claude Opus 4.6

19 ARTICLES

last mo.

OpenAI Releases GPT‑5.6 and GPT‑Live After U.S. Security Review

6 ARTICLES

last mo.

White House Sets Voluntary Pre-Release Review for Frontier AI

7 ARTICLES

last mo.

Anthropic Restores Fable 5 but Safety Filters Quietly Route Many Calls to Opus 4.8

14 ARTICLES

last mo.

U.S. Lifts Export Controls on Anthropic’s Claude Models

Model Development

OpenAI Pauses Frontier Training After AI Agent Escaped Sandbox

Top 1% of U.S. Firms Spend a Median $7,400 Per Employee on AI, Ramp Finds

Anthropic Says Claude Agents Turned on Each Other in Simulated ‘Turf War’

NVIDIA Releases Nemotron 3.5 Lightning and NeMo Switchyard

OpenAI Pauses Astra and Releases GPT‑5.6‑Cyber to Vetted Security Partners

OpenAI Loses Head of Ethics Less Than a Year After Hire

Alibaba Publishes Qwen 3.8–Max and Will Release Open Weights Next Week

Lilian Weng Leaves Thinking Machines Lab and Rejoins OpenAI to Lead Recursive Self-Improvement Research

U.S. Opens Inquiry Into Chinese AI 'Distillation' as Xi Plans September 24 Washington Visit

Moonshot’s Kimi K3 Prompts U.S. Accusations of Model Theft

Never miss stories about

Model Development

Moonshot Unveils Kimi K3, a 2.8‑Trillion‑Parameter Open‑Weight AI Model

OpenAI Builds GPT-Red to Hunt Prompt‑Injection Flaws

OpenAI’s GPT-5.6 Sol Reported to Delete Files and Production Databases

OpenAI Positions GPT-5.6 Sol as Cheaper, More Token‑Efficient Rival to Anthropic’s Fable

Anthropic Extends Free Claude Fable 5 Access for Paid Plans to July 19

Anthropic Finds Hidden ‘J‑Space’ That Reveals Near‑Future Tokens Inside Claude Opus 4.6

OpenAI Releases GPT‑5.6 and GPT‑Live After U.S. Security Review

White House Sets Voluntary Pre-Release Review for Frontier AI

Anthropic Restores Fable 5 but Safety Filters Quietly Route Many Calls to Opus 4.8

U.S. Lifts Export Controls on Anthropic’s Claude Models

Never miss stories about

Model Development