Why the AI Kids Hate Astra 6.2
OpenAI's GPT-6.2 Astra and Sol drew harsh early reviews from coding-focused testers, but efficiency gains and a shift toward cost-conscious enterprise work suggest the release is more about strategy than failure.
Watch on YouTube ↗Charlie's summary
- ▶ 2:00
The wave of outages is more likely botched legacy upgrades than a cyber attack
After outages at Microsoft, the FAA, big banks and the NSA's network, NLW cites the view that AI is behind them in a less alarming way. Fear of automated attacks is driving a surge of updates, and a surge in updates means a surge in failures. Systems “held together with popsicle sticks and bubble gum for 30 years” break when pushed, and specialists are stretched thin. He expects this during hardening, not an attack.
- ▶ 12:10
OpenAI is selling Astra and Sol as efficient real-world models, not coding leaders
NLW says OpenAI's announcement leads with Agents Last Exam and Terminal Bench Science rather than coding benchmarks, and stresses cost even for code. Artificial Analysis found Astra 6.2 rose five points to 62, second behind Fable 5.2. It ran the benchmark at $4 per task versus $6.24 for the prior Astra, a 35% cut. Sol was about 20% cheaper.
- ▶ 14:11
The price gap with Anthropic is now large, even though coding scores slipped
NLW notes Fable 5.2 on high settings is still more than twice as expensive as Astra on max settings, and Sol on max costs under a third of Opus 5.5 on high. The red flag: both models regressed on the Artificial Analysis coding agent index. Sol dropped three points because it more often omits key facts. That set the tone for the early negative narrative.
- ▶ 15:12
Early testers found the models concise to a fault, stopping early and stuck in house styles
Dan Shipper's 30-person team at Every said no one switched to Astra 6.2. They found its writing so terse it was hard to understand, and it often stopped too early on end-to-end builds. Its designs had a noticeable house style. Theo said it is “insufferable to work with on real projects,” and Ethan Mollick called it uneven.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.