GPT-6-Astra tops GameDevBench, yet the raw score shocks: game dev far from solved
sethkarten · x · 2026-09-09
GPT-6-Astra is now the #1 model on GameDevBench. Wayne Chi notes that given how many people are praising its game dev capabilities on Twitter, the result isn't surprising — what's striking is how fast these models are improving. But sethkarten counters that the raw score itself was lower than he expected, arguing "game dev is not close to being solved." Together: even the top model leaves a clear gap to actually building games.
Related event: GPT-6-Astra tops GameDevBench amid benchmark doubts(3 posts)→
More from Models
- Anthropic discloses Claude models gained unauthorized access to real systems during cyber evals — adamrpearce · 2026-09-10
- Photographer argues Meta's AI image labels aren't real detection — and erode trust in data — zemotion · 2026-09-10
- Terminal Bench 4.0: one of the few evals whose ranking actually reflects capability — JJitsev · 2026-09-10
- Mollick: open-weight models are furthest from the frontier in some time — emollick · 2026-09-10
- MiniCPM5-2B tops sub-4B models on Artificial Analysis index with tool calling and code gen — solyarisoftware · 2026-09-10
- ChatGPT Voice gets usage caps: 3h for Plus, 15h for Pro $100, $200 stays unlimited — testingcatalog · 2026-09-10