OpenVLHarness turns any LLM into a visual reasoner, claiming +9.9 points across 22 datasets
hsu_byron · x · 2026-10-21
cfyang58 released OpenVLHarness, a framework that turns any LLM into an evidence-based reasoner that thinks and codes in visual space.
- Claims GPT-6 Astra gains +9.9 points on average across 22 datasets when hooked up.
- Beats Codex and other VL harnesses by 7.7–8.5 points at ≤37% of their cost.
- Early-stage claims; details are in the follow-up thread, no third-party replication yet.
More from Models
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21
- Muse Glimmer model card leaks: 80.0 on AA-LCR long-context eval, beats Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21