Reddit user reports Astra regression: fake website links, misread prompts, higher usage
median0rc · reddit · 2026-09-09
A user who relies on ChatGPT for math and ML work says that despite Astra's promising benchmarks, real-world use showed regressions: recommending fake websites, higher consumption than advertised (compared to Sol), prompt misunderstandings unseen in years, and inconsistent improvements. The post is anecdotal and asks whether others see the same.
Related event: GPT-6 Astra launch: dazzling demos clash with rocky real-world tests(10 posts)→
More from Models
- Anthropic's Terra returns with better writing and 2X usage vs 6.1 Sol — weswinder · 2026-10-27
- User catches model failing to parse 400-paper file, citing stale news sources instead — jungofthewon · 2026-10-27
- Odd eval result: Gemini 4.1 Flash gets more accurate when asked to control its CoT — dfrsrchtwts · 2026-10-27
- Poster surprised that v4.1 flash gets more accurate when asked to control its CoT — dfrsrchtwts · 2026-10-27
- Dev claims Anthropic is baking 'sabotage' into model behavior for 'wrong' engineering — cephaloform · 2026-10-27
- Gemini fails at listing unread emails, tells user to search Gmail manually — gethackteam · 2026-10-27