Seth Frame (@FrameSeth) profile picture
Joined March 2026
Replying to @sawyerhood
Polymarket says no to spud. Here's 4.6 tho
1
766
The Mythos card fires quite a few shots at Grok lol
80
Replying to @Angaisb_
some are slightly benchmaxxed, but even on proprietary benchmarks that still outperform the new GPTs
1
206
Replying to @scaling01
Yeah same for Topology Proving. It's the best american small model. It's a terrible small model globally
4
1,202
GPT released two new models and I ran them on TopoBench (Lean-verified Topology). Nano beats Flash-lite at same price. Kimi and Deepseek beat Mini handily at a lower price. Best cheap American models, not best globally for math proving.
56
TopoBench, a new Lean 4 verified topology benchmark. 100 formally verified theorem-proving items generated by DeepSeek v3.2, Gemini 3.1, GPT-5.3 Codex, and Claude Opus 4.6. No current frontier model cracks 60%. GPT 5.4 lower than Opus 4.6
128