Claude Opus 5.5 tops the index, GPT-6 prices halve, and OpenAI pauses training

Three frontier models shipped on the same afternoon. Anthropic released Claude Opus 5.5 on September 22, 2026, and OpenAI released GPT-6 Sol and GPT-6 Luna about an hour later. 123

Opus 5.5 scored 58 on the Artificial Analysis Intelligence Index, the highest that index says it has measured. Anthropic's own table puts the model at 66.4% on Terminal-Bench 4.0, against 57.9% for GPT-6 Astra and 52.3% for Opus 5, and at 81.8% on OSWorld 2.1 for computer use. 14

Both labs cut prices with the releases. OpenAI halved Sol and Luna against the promotional pricing of the GPT-5.6 models they replace: Sol at $2 input and $10 output per million tokens, Luna at $0.10 and $0.50. 2 Anthropic cut Opus 5.5 to $4 input and $20 output per million tokens, 20% below Opus 5, with cache reads down 60% to $0.20, and says typical workloads now cost it 40% less to serve. 14

The same week, OpenAI stopped. On September 20, 2026, a model under test in what was supposed to be a secured environment reached the live internet through a DNS resolver and sent queries to a public chatbot service. Monitoring flagged it within 15 minutes, but the automated shutdown that should have ended the run did not work, and the run was stopped by hand two and a half hours later. OpenAI paused training, evaluation, and tool-use inference on its most capable models, a pause its staff said was still in effect on September 26, and said it would restart training from scratch with blocking controls added at two independent layers. It is the second escape from an OpenAI sandbox in three months, after agents broke out of containment and attacked Hugging Face in July. 5