OpenAI's GPT-6 Sol and Luna and Anthropic's Claude Opus 5.5 bring the latest model advances into daily professional work.Their emphasis on reliability, speed, and more natural communication gives businesses new reasons to revisit what they ask AI to do.
OpenAI introduced GPT-6 Sol and Luna, extending advances from its flagship Astra model into faster options for professional work, coding, and computer use, where AI operates software to complete tasks. Sol is positioned as a workhorse, with Luna serving as the smaller option. Astra remains the choice for the most demanding projects.
OpenAI reports that Sol made about half as many factual mistakes as its predecessor on an internal test built from conversations where users had flagged errors. That is a specific evaluation, not a prediction that every task will occur this way. Sol and Luna launched in ChatGPT Work and Codex for paid users; Free and Go users can access Luna in the desktop app.
Anthropic's Claude Opus 5.5 performs at the level of Fable 5.1 on most work, according to the company. Anthropic reports output more than 30% faster than Opus 5, and early testers found its writing clearer. The company is also raising usage limits on paid plans, giving people more room to work with the models.
SmarterX founder and CEO Paul Roetzer examined what this generation of releases means for professionals, and why familiar-looking upgrades still matter, on Episode 243 of The Artificial Intelligence Show.
2 - New GPT-6 models joining Astra: Sol and Luna
About half - The reduction of Sol's factual mistakes versus its predecessor on OpenAI's internal evaluation
>30% - Opus 5.5's output speed improvement versus Opus 5, according to Anthropic
40% - Anthropic's estimated typical-workload cost reduction versus Opus 5 at default settings
A familiar experience but more useful. Roetzer compares these releases to an iPhone upgrade. The product looks familiar, but the improvements make it more useful. In his assessment, "They're just more reliable, more efficient, more aligned." Alignment refers to how well a model follows intended instructions and boundaries.
That combination matters when someone uses AI repeatedly throughout a workday. Fewer factual errors can mean less rework. Faster output can make it easier to iterate. Clearer writing can make the result easier to understand and assess. Each improvement is worth testing against the work an employee actually needs to finish.
Let AI do it for you. Roetzer sees the companies increasingly explaining their models through the work of lawyers, accountants, consultants, marketers, and salespeople. As he puts it, "It's more about, 'Here's how you can use it in your day-to-day work.'" He expects more customization for particular industries, functions, and roles.
Safety is becoming a bigger part of new models, companies say. Anthropic says Opus 5.5 underwent external evaluation before release and achieved its strongest results yet on its automated behavioral audit, a test of how models behave across simulated situations. Roetzer expects safety and alignment to receive more attention as the models become more capable. It's notable that these are company test results and not proof that failures have been eliminated.
Choosing a model still requires judgment. Roetzer used Fable for demanding prototype development and Sonnet for thinking through ideas. The choice depended on what he needed from each interaction. New releases give professionals more options, but learning to direct the work remains essential.
"Train the user first, and then use the model that's going to give you the best chance to create the output you're seeking to create."
— Paul Roetzer, founder and CEO of SmarterX on Episode 243 on The Artificial Intelligence Show
It's worth revisiting a task that an earlier model handled poorly. Give the new model the same context, define what a good result looks like, and compare the work. Check factual accuracy, clarity, and the amount of intervention required. That makes a release meaningful in ways a team can evaluate.
Model selection should become part of AI education. Employees need practice deciding when a task calls for the strongest model and when a faster option is sufficient. Keep examples of successful work and the instructions that produced it, so an individual experiment can improve how the broader team works.
Everyday use will test the companies' claims. The useful evidence is whether these models produce better work with fewer needed corrections across tasks professionals repeat every day. Roetzer also expects more emphasis on adapting models to specific roles. That would make the quality of a model's fit for a job increasingly important alongside its general capabilities.
Integrating AI tools into existing workflows is the most-requested training topic, selected by 58% of respondents in the 2026 State of AI for Business Report. That demand fits the challenge these releases create: employees need help deciding where AI belongs in their work and how to use it well, even as the available models improve.
Based on more than 2,100 responses from professionals across roles, functions, and industries, the report examines adoption, training, and organizational readiness. It provides a useful starting point for comparing an organization's education priorities with what professionals want to learn. Read the full report →