Sol and Luna bring GPT-6’s prices down
New models arrive in Work and Codex as OpenAI and Anthropic compete on cost. What changes—and what still needs checking.
OpenAI released GPT-6 Sol and Luna on September 22, making its newer model family more accessible for everyday work. The immediate story is cost: Sol’s standard API input and output rates are half its predecessor’s promotional rates. TechCrunch’s launch report places the release alongside Anthropic’s Opus 5.5 announcement, just 90 minutes earlier.
For ChatGPT users, the destination matters. OpenAI says Sol and Luna are rolling out in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users get Luna in the desktop app. They are not yet available in Chat, and availability may take time to appear during the rollout. OpenAI’s announcement
Developers get a more concrete comparison. Standard API pricing per million tokens is $2 input and $10 output for Sol, and $0.10 input and $0.50 output for Luna. These are token rates, not fixed prices for completing a task. Tool calls, processing options, long prompts and how much work the model does can change the bill. They also do not establish a reduction in ChatGPT subscription prices.
OpenAI reports fewer factual mistakes and better coding behavior. Its internal factuality test found roughly half as many mistakes from Sol as from its predecessor. But that evaluation uses conversations selected because users had flagged errors; OpenAI explicitly says they are not representative of typical use. Treat the improvement as a reason to test familiar tasks, rather than permission to stop checking answers. Evaluation details
The competition is making similar promises. Anthropic says Opus 5.5 costs about 40% less on its typical-workload tests, while its input and output token prices fall 20%. That distinction matters: efficiency and sticker price are different measures. The Verge reports that the release also brings stronger safeguards, including routing some sensitive requests to earlier models. Neither company’s announcement is an independent head-to-head verdict.
There is a policy test alongside the product test. On September 21, OpenAI proposed international technical standards covering measurements, incident reporting and human review. The proposal explicitly separates those standards from licensing or mandatory prerelease approval. Shared definitions could help, but they do not by themselves establish enforceable oversight.
Our practical recommendation: try one repeatable task with an answer you can judge. Compare the finished result, correction time and total cost with your existing setup. Keep consequential actions subject to review. Cheaper iteration is useful; reliable work is the outcome worth paying for.