FeedSources
OR

OpenAI Research

308 articles total

Go to source

We believe our research will eventually lead to artificial general intelligence, a system that can solve human-level problems. Building safe and beneficial AGI is our mission.

Go to source
  •  Scientific computing in the age of agentic AI
  •  GPT-Red: Unlocking Self-Improvement for Robustness
  •  GPT-5.6: Frontier intelligence that scales with your ambition
  •  GPT-5.6 System Card
  •  Separating signal from noise in coding evaluations
  •  Introducing GPT-Live
  •  GPT-Live System Card
  •  Introducing GeneBench-Pro
  •  Previewing GPT-5.6 Sol: a next-generation model
  •  GPT-5.6 Preview System Card
  •  A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry
  •  Introducing LifeSciBench
  •  Dreaming: Better memory for a more helpful ChatGPT
  •  Introducing new capabilities to GPT-Rosalind
  •  Strengthening societal resilience with Rosalind Biodefense
  •  An OpenAI model has disproved a central conjecture in discrete geometry
  •  Advancing voice intelligence with new models in the API
  •  GPT-5.5 Instant System Card
  •  GPT-5.5 Instant: smarter, clearer, and more personalized
  •  Where the goblins came from
  •  Introducing GPT-5.5
  •  GPT-5.5 System Card
  •  Introducing OpenAI Privacy Filter
  •  Introducing ChatGPT Images 2.0
  •  Introducing GPT-Rosalind for life sciences research
  •  Inside our approach to the Model Spec
  •  How we monitor internal coding agents for misalignment
  •  Improving instruction hierarchy in frontier LLMs
  •  GPT-5.4 Thinking System Card
  •  Introducing GPT-5.4
  •  Reasoning models struggle to control their chains of thought, and that’s good
  •  Extending single-minus amplitudes to gravitons
  •  GPT-5.3 Instant: Smoother, more useful everyday conversations
  •  GPT-5.3 Instant System Card
  •  Why SWE-bench Verified no longer measures frontier coding capabilities
  •  Our First Proof submissions
  •  Advancing independent research on AI alignment
  •  Introducing EVMbench
  •  GPT-5.2 derives a new result in theoretical physics
  •  Scaling social science research