Commenters welcomed GPT‑6.2 Astra and Sol as potentially strong upgrades, especially if their benchmark gains and low prices translate into better task-completion value. Some cited updated results placing both models near the intelligence-cost frontier; one tester found Astra clearer, faster, and more practical than its predecessor, while another praised its ability to build a polished 3D game. Others liked Sol as a dependable workhorse, and a business owner said OpenAI’s integrated tools offered more value than competing products. Several also hoped the release would improve on earlier Astra’s tendency to over-engineer simple software tasks.
The divide was over whether benchmark results match everyday performance. Some commenters questioned the usefulness of familiar visual tests and wanted comparisons on real work; others reported poor office-document examples, unwanted changes beyond the requested scope, regressions in reliability, or excessive latency that made Sol impractical. One user’s early test suggested Astra had improved, while another found it still ignored instructions. Commenters also compared the models unfavorably or favorably with Claude, Gemini, and Fable, reflecting different experiences and preferences. A few raised broader concerns: AI systems refusing technical security and reverse-engineering tasks, and whether models become less capable around release time. Overall, enthusiasm about cost, capability, and product integration coexisted with skepticism about usability, instruction-following, and real-world quality.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.