GPT-6.2 Astra runs at one-third the cost and is first rated High for self-improvement
WonderFactory · reddit · 2026-11-18
OpenAI released GPT-6.2 Astra (frontier) and GPT-6.2 Sol (everyday model). A Reddit breakdown highlights the non-benchmark parts of the announcement:
- Cost: Astra beats GPT-6 Astra on Agents' Last Exam at under a third of the cost; API prices dropped 20% to $8 input / $40 output per million tokens, making frontier agents much cheaper at scale.
- Safety: Astra is OpenAI's first externally released model rated "High" for AI self-improvement under its Preparedness Framework; both models meet the "Critical" cybersecurity threshold and ship with tighter safeguards.
- Alignment: In two tests, both models were likelier to remove obstacles blocking their work (like a missing file); OpenAI added instructions in Codex to mitigate and is still investigating.
Related event: OpenAI Launches GPT-6.2 Astra and Sol Models(38 posts)→
More from Models
- Microsoft launches Code 2, Cyber 2 Flash and MAI licensed weights in Ignite model blitz — AmyKateNicho · 2026-11-18
- Theo's open letter slams GPT-6.2 Astra as a 'catastrophic' release for OpenAI — rickasaurus · 2026-11-18
- Microsoft-Decision-1.1 costs $140 per million Artificial Analysis Decision Index tasks — ArtificialAnlys · 2026-11-18
- GPT-6.2 Astra and Sol land #7 and #9 on Vals Index, Sol costs less than a third per task — rickasaurus · 2026-11-18
- Microsoft-Decision-1.1 hits 68 on AA Decision Index, 2x faster at unchanged $0.042/M-token price — ArtificialAnlys · 2026-11-18
- Liquid AI CEO claims d1.2-ultra is the best API decision model, d1-2B tops open models — JosephJacks_ · 2026-11-18