| Oct 15, 2026 | Released ORL (paper, code, models): diffusion language models that decode a single request at up to 2,788 tokens/s on one RTX 5090 and 7,708 tokens/s on 8× B200 |
| Sep 30, 2026 | Released QATFactory: a deployment-aligned framework for quantization-aware training and distillation of LLMs |
| Sep 15, 2026 | Paper I-DLM selected as an Oral Spotlight at COLM 2026 (24/856 accepted papers, out of 2900+ submissions) |
| Aug 21, 2026 | Paper Osprey got accepted to the EMNLP 2026 Main Conference (2,719/17,669 submissions; 15.4% acceptance rate) |
| Jul 08, 2026 | Paper I-DLM got accepted to COLM 2026 |
| Apr 15, 2026 | Released I-DLM: the first diffusion language model to match AR-thinking model quality, achieving 69.6 on AIME-24 with 2.9–4.1x throughput |
| May 01, 2025 | Joined OpenAI as a Member of Technical Staff Intern — “The first batch of interns in OpenAI’s history.” |