Photo by Kindel Media on Pexels OpenAI released GPT-6.2 Astra on November 17, 2026 alongside a sibling model, GPT-6.2 Sol — and with a 20% price cut from the previous Astra generation. The new API rate is $8 per million input tokens, $40 per million output tokens , down from $10/$50. Cached reads drop to $0.40/million. On paper, it's a compelling upgrade at a lower cost. Within 24 hours of launch, prominent developers were calling it "the worst model OpenAI has ever released." Both things are true simultaneously. The price cut is real. The backlash is real. Here is what actually matters depending on how you plan to use it. The Numbers at a Glance Model Input ($/1M) Output ($/1M) Cached Input Context GPT-6 Astra (prior) $10.00 $50.00 $0.50 ~1M tokens GPT-6.2 Astra (new) $8.00 $40.00 $0.40 ~1M tokens ...
Photo by Google DeepMind on Pexels OpenAI shipped GPT-6.2 Astra and Sol on November 17. The headline is a 20% price cut on Astra's API and a Sol that nearly keeps up with Astra on coding at one-fifth the cost. The sub-headline, buried in user reports: power users are calling Astra "essentially unusable" and "uneven and weird." Both things are simultaneously true, and knowing which applies to your workload matters before you migrate any production code. The Numbers That Actually Matter Model Input / Output (per 1M tokens) FrontierCode 2.0 Terminal-Bench-Science Data Analysis (Plotly run cost) GPT-6 Astra (prior) $10 / $50 — ~72% $19.97/run GPT-6.2 Astra (new) $8 / $40 — 77–78% $4.11/run (95% accuracy) GPT-6.1 Sol (prior) $2 / $10 39.0% 57.0% $2.61/run ...