Summary
OpenAI has released two new models, GPT-6 Soul and GPT-6 Luna, and this video from Matthew Berman walks through what changed. The launch landed the same day as Anthropic’s Opus 5.5 and one day after xAI’s Grok 4.7, making it a crowded week for frontier releases.
Pricing is the headline. Soul is listed at $2 per million input tokens and $10 per million output tokens, a 50% cut from GPT-5.6 Soul. Luna drops a further 50% on top of its earlier 80% reduction, landing at 10 cents per million input tokens and 50 cents per million output tokens. The video argues Luna is a workhorse model that can handle the large majority of everyday tasks.
The review then looks at benchmark results, including Automation Bench, Frontier Code and OSWorld, plotting score against cost per task. Luna Max reaches 66.6% on Frontier Code at about 22 cents per task, while on OSWorld it trails GPT-6 Astra, the top OpenAI model, by roughly 53% to 73%. A side-by-side chart against Opus 5.5 shows performance gaps of only six to seven points but large price differences.
Both models are available in ChatGPT, Work and Codex for paid tiers starting today, with Luna also reaching free and Go users in the desktop app. The video closes with commentary on why both OpenAI and Anthropic are competing on efficiency and price.
📺 Source: Matthew Berman · Published September 23, 2026
🏷️ Format: News Analysis
