GLM-5.3 Flash trails full model by 5.6 points at 17x lower cost
Together AI ran 900 DeepSWE rollouts on GLM-5.3 and GLM-5.3 Flash. Flash trails by 5.6 points at pass@1 but only 2.6 at pass@4, at 17x lower cost ($0.24 vs $3.99 per rollout). A cascade starting with Flash solves 80.9% of tasks at $1.70 each.
How this story unfolded
2 days · 2 reports · 3 community posts · from Aug 28
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Ksyon robot demo shows pseudo-realistic head movements and local vision
- AI models to become 100x faster, enabling 300 civilizations in the time of 3
- Hermes with Claude sub teased
- Tool extracts structured data from receipt images using LLMs or OCR
- Midjourney user shares Shadowrun-inspired art set