15M-param agent beats Craftax boss 44% of time for a cent; GPT-6 Astra wins 1% for $83
hyperparticle · x · 2026-10-08
rekursiv.ai showed a small neural net can crack the hard Craftax benchmark without an LLM: their agent meta-harness trained a 15M-param agent in 3 weeks that beats the final boss 44% of the time, at a cost of one cent and 10 seconds per run.
By contrast, LLM-powered GPT-6 Astra won only 1% of the time, taking 24 hours and $83. A striking counter-intuitive cost-vs-capability result.
Related event: 15M-Parameter Tiny Model Beats GPT-6 on Craftax for One Cent(2 posts)→
More from Research
- OpenAI's new result proves 2005 edit-distance embedding optimal; researcher distills proof to 2.5 pages with AI help — thegautamkamath · 2026-10-08
- Experiments show smarter models and higher effort write better LLM-judge evals — danshipper · 2026-10-08
- Tencent's WorkForge scales verifiable training environments for long-horizon work agents — teortaxesTex · 2026-10-08
- Masked Geometric Encoder boosts 3D foundation models via frame dropping and self-distillation — zhenjun_zhao · 2026-10-08
- DensiTok: flow-matching token densification lets frozen feed-forward 3DGS see unseen views — zhenjun_zhao · 2026-10-08
- Warping flat-port views into pinhole perspective for underwater 3D reconstruction — zhenjun_zhao · 2026-10-08