Two announcements separated by an hour and a half: Anthropic releases Opus 5.5, OpenAI responds with GPT-6 Sol and GPT-6 Luna, billed half as much as the 5.6 series.
On September 22, OpenAI released two new models of the GPT-6 family. Sol targets long tasks and code; Luna, lighter, is designed for volume: summaries, information extraction, short queries. The company claims about half as many errors for Sol as its predecessor, with a reliability comparable to that of Astra but at a much lower cost, thanks to gains on cache and inference.
On the pricing side, the change is clear: both models are priced at about half the price of the 5.6 series. Luna goes down to $0.10 per million tokens in input and 0.50 in output, making it the cheapest option in the range. Sol and Luna can be accessed in ChatGPT Work, in Codex and via the API; Luna is also open to free and Go accounts on the desktop app.
The timing is not innocent. Anthropic had released Opus 5.5 about ninety minutes earlier, with its own price drop. OpenAI claims that Sol and Luna are ahead of competitors in its internal tests — a comparison that no third party has yet to replay.
Read the full article on the original source
Go to the source →