From GPT-5.0 to GPT-6.0 Astra: one year of model progress, documented via the same game prompt
mt229 · reddit · 2026-09-05
A blogger documented a year of OpenAI model progress — GPT-5.0-high through GPT-5.2-codex, GPT-5.5-high, GPT-5.6-sol-high, and now GPT-6.0-astra-high — by prompting each version to build the same browser MMORPG from a single prompt. Older versions stay playable online (bugs intentionally preserved), making the capability jump between generations directly visible. Tested on Windows 11 + Chrome only; the play link is hidden in the images.
More from Models
- Inference-Time Scaling: More Compute, No Retraining, Better Answers — _jaydeepkarale · 2026-09-05
- RL Hill-Climbing Creates Model Spikes, So Judge AI With Multiple Models — dejavucoder · 2026-09-05
- TheZvi Polls Readers on Their Most-Used AI Model Across All Tasks — TheZvi · 2026-09-05
- Extropic releases first transformer family built for probabilistic thermodynamic hardware — Mountain_Cream3921 · 2026-09-05
- NVIDIA's NVFP4-quantized Qwen3.8-Flash-Next image-text-to-text model trends on Hugging Face — nvidia · 2026-09-05
- Developer complains Astra's safety guardrails block even basic tasks — jarrodwatts · 2026-09-05