Qwen 3.8 27B scores 34 (99th percentile) on ACT run locally with vision
on_line187 · reddit · 2026-08-20
A Reddit user had Qwen 3.8 27B Instruct (Q80 GGUF, LM Studio, 2× RTX 3090 full offload, 32k context) take two official ACT practice tests — 342 questions fed as raw PDFs to test vision, with no human help and no retries.
Results: composite scores of 36 (the maximum) and 34, overall accuracy 95.3% (326/342), zero blanks. Reading was perfect on both papers (72/72); math near-perfect; English and science 33-35. A 34 is roughly the 99th percentile.
It took 177 minutes for both tests (88 min each) versus 165 minutes allowed per test for humans — slower than expected because the model read questions visually from PDFs rather than plain text.
More from Models
- Meta open-sources Muse Spark Open agent model with day-0 TPU support in SGLang — BanghuaZ · 2026-10-16
- Saying "load-bearing" in your prompt makes Opus 5.5 stop holding back — zacharynado · 2026-10-16
- Americans get their Claude accounts banned while on business trips in China — dotey · 2026-10-16
- I priced every Grok Bot call: $20 to $200 plan in 5 days, long jobs eat 52% — jonathanmalkin · 2026-10-16
- NVIDIA releases Nemotron-3.5-Embed-1B, a multimodal embedding model covering 34 languages — jacek2023 · 2026-10-16
- EpochAI pegs GPT-6 training at up to 1.8e27 FLOP; scaling01 calls param estimates way too high — scaling01 · 2026-10-16