OpenAI GPT-5.5 Instant Release and Benchmarks
2 videos · score: 5,293 · first seen Jun 9, 2026
OpenAI has released GPT-5.5 Instant, showcasing improved performance on the new DBSE coding benchmark with a 70% success rate, while creators highlight both its strengths—such as reduced hallucinations and near-top-tier performance in specialized tasks—and weaknesses, like poor refusal rates against adversarial prompts, making it a hot topic in AI circles.
AI code benchmarks lied to us
Data Curve's new DBSE benchmark shows OpenAI's GPT-5.5 outperforming Anthropic's Opus at 70% vs 54%, while criticizing older SWEBench Pro as contaminated and unrealistic.
OpenAI's ChatGPT 5.5 Instant: The Good, The Bad And The Insane
OpenAI released ChatGPT 5.5 Instant, which matches top thinking models on cybersecurity and biology tasks with halved hallucination rates, but Dr. Koa Eher criticizes its weaker model-level refusal against adversarial prompts as patched by external classifiers rather than fundamentally fixed.