Open-source Tessera-27B thinks in compressed CJK, cutting agent thinking tokens by half
ghgi_ · reddit · 2026-10-15
Zellige AI open-sourced Tessera-27B-B-CJK (Apache-2.0), a Qwen3.8-27B finetune that thinks in compressed CJK shorthand while emitting plain English and code. On 11 held-out coding-agent tasks it uses 0.50-0.64x the base model's thinking tokens per turn, with 100% valid tool calls and zero repetition loops in 88 matched turns. Trained with just four LoRA steps on Tinker using the base model's own outputs; weaker on competition math. Weights, LoRA, GGUF and NVFP4 available on Hugging Face, with a paper covering the pipeline and failures.
More from coding & agent
- Isolating Antigravity chats per subfolder with AGY Rules and PreInvocation Hooks — bibryam · 2026-10-16
- Gemini shown off modding Minecraft to turn players into any Spider-Man — zacharynado · 2026-10-16
- Workflow: annotate UI screenshots in tldraw, hand to Claude, get better UI — max__drake · 2026-10-16
- Beyond chat wrappers: designing stateful, multimodal AI apps as interaction conductors — dr_cintas · 2026-10-16
- IBM engineer maps 5 levels of AI coding, from manual to a 3-day agent software factory — TejasKumar_ · 2026-10-16
- 10 open-source repos that let your AI agent reverse-engineer apps and games — victor_explore · 2026-10-16