User has GLM 5.3 Flash research game one-shots: 145 calls, 15M tokens in 48 min
yeah_likerage · reddit · 2026-09-11
Tired of gaming one-shot benchmarks, the author asked a self-hosted GLM 5.3 Flash to research the history of those benchmarks and extrapolate when AI-generated modern games become playable.
- Wall time 48 minutes, 145 model calls
- 15M input tokens, 202K output tokens
- Home-hosted GLM 5.3 Flash Max with Hermes
More from coding & agent
- Economist replicates an academic paper with a research agent, 'almost zero' manual work — soumitrashukla9 · 2026-09-11
- Warp exec runs six non-engineering teams like engineering, all on Claude Code — round · 2026-09-11
- Automating content creation: build the research pipeline first, writing comes last — EXM7777 · 2026-09-11
- Anthropic engineer on self-improving agents, and why multi-agent workflows should be graphs, not lines — Aiden_Tech_Ai · 2026-09-11
- Muse can generate a podcast on any topic; dev builds full series via MCP — RichardsonDx · 2026-09-11
- My agent spent $3.64 answering one question I thought was free — GoldBroccoli7073 · 2026-09-11