Anthropic further claims that the savings are closer to 40 percent compared to Opus 5 for typical workloads with the default configuration, because in addition to the token cost decreasing, Opus 5.5 also uses fewer tokens when completing tasks.
Opus 5.5 is said to be remarkably capable in “high-risk areas” like cybersecurity and biology, so the same protections that applied to Fable 5.1 will also apply here: your requests can be automatically and transparently routed to an older model if they are flagged as entering protected territory.
GPT-6 Sun and Moon: a modest, cost-focused upgrade
OpenAI’s release of GPT-6 Sun and Moon is an iterative step forward. Not long ago, the company introduced its Sol, Terra, Luna naming convention for its GPT-5.6 family of models. And more recently (just this month), it introduced the GPT-6 Astra, which has been its most advanced and powerful model.
It may not be obvious what those names mean, so here’s a quick summary. Astra is the most aggressively powerful (and expensive) model, designed for coding, research, etc. Sol is a capable but more efficient and affordable alternative, designed to be the kind of daily driver for tasks like that. Terra is the balanced and general purpose model. And Luna is the quick and economical option.
You could reasonably and roughly position Astra against Anthropic’s Fable, Sol against Opus, Terra against Sonnet, and Luna against Haiku, but it would be an imperfect mapping, especially for the high-end models.
OpenAI says that GPT-6 Sun and Moon were trained with similar methods to those used to train GPT-6 Astra. Depending on the benchmark, they are sometimes a few percentage points more capable than their predecessors at certain tasks, but cost half as much to use.
