Gael says OpenAI has publicly described internal models stronger than Astra. His prediction: compute and product work are the hold-up, and the leaderboard could change again within weeks.
Gael's read: GPT-6 Sol and Luna are cheaper to run, but not clearly smarter than the prior models. The 50% higher usage limit matters if you hit your cap; otherwise, this release may feel like a step back.
Gael ran Opus 5.5 in Claude Code for 2.5 hours on the $20 plan and used just 8% of his weekly limit. After years of criticizing Anthropic's limits, he calls this its best model release yet.
Could AI computer use automate 80% of desk work? That's Gael's prediction, not a measured result. The shift he's watching: AI operating browsers, editing tools, dashboards and design apps.