Tess-4.5-27B open-sourced: same scores with 25% fewer tokens than its Qwen base
jacek2023 · reddit · 2026-10-20
migtissera (author of Synthia) released Tess-4.5-27B, an open-source Qwen finetune (GGUF on Hugging Face), pitched as "less thinking, same answers." Across all 150 BenchLocal scenarios it hits the same scores with 25% fewer tokens than its base (262K vs 349K); on CLI-40 it uses 32% fewer tokens while scoring higher, and 43% fewer on data extraction. Fewer tokens mean faster answers, less context spent, longer sessions — compounding with a draft head on local GPUs.
More from Models
- vLLM ecosystem ships d3-turbo: multimodal decision model, 15x faster, tops Jev Decision Index — vllm_project · 2026-10-20
- MOSS 5B architecture diagram built from official code, MIT-licensed — TheLatentExplorer · 2026-10-20
- Devs say no model loyalty: will switch to OpenAI for better limits — Angaisb_ · 2026-10-20
- reuF2F benchmark: 244 REU math problems formalized in Lean, best model solves only 109 — wellecks · 2026-10-20
- Andrew Ng: AI math breakthroughs are largely not from mathematicians uploading ideas — andrewgwils · 2026-10-20
- Pangram 4.1 fixes AI-detection false positives after community pushback — GolinoHudson · 2026-10-20