Industry
We pointed Meta-Zenith at one task: make Qwen3.8-27B-NVFP4 faster on a single RTX 5090. 111 self-directed trials later, no human in the loop…
We pointed Meta-Zenith at one task: make Qwen3.8-27B-NVFP4 faster on a single RTX 5090. 111 self-directed trials later, no human in the loop: +63.5% throughput (geomean, 18 settings) 4.1× at 65K conte
We pointed Meta-Zenith at one task: make Qwen3.8-27B-NVFP4 faster on a single RTX 5090. 111 self-directed trials later, no human in the loop: +63.5% throughput (geomean, 18 settings) 4.1× at 65K context tool-call quality unchanged https://github.com/Intelligent-Internet/Qwen3.8-Inference-MetaZenith
Related
- This blog is a great review of the current environment on self-improvement We leveraged a number of things here for our harness of harnesses…
- We have seen multi model harnesses for cheaper & faster tasks What about for the hardest challenges? What about open source? Proud to share …
- Follow & contribute to Zenith in our repo Meta-Zenith and other updates to come https://github.com/Intelligent-Internet/zenith
- It seems @OpenAI has cracked recursive self-improvement for Codex
- New research: long-running agents often fail by stopping too early, not because the model can't make progress. We tested 5 harness designs a…
Source: Emad Mostaque (X) | 2026-09-09