basal-omni-2.0-mini: an open typed decision model with two-way reading, 0.811 on Werdykt
rkinas · reddit · 2026-11-14
The author released basal-omni-2.0-mini, an early small version of basal-omni 2.0 built on Gemma 4 E4B. Key points: two-way reading of the state before answering (needs basal engine 2.5+); scores 0.811 on Werdykt, a hidden 5,000-decision Polish/English benchmark; outputs a calibrated probability in a single forward pass, strictly confined to the given options with no free-text generation; supports choice/noul/score types (plus multi/act/tool in the engine) with option codes up to 255. Model and inference engine are both open-sourced on Hugging Face and GitHub.
More from Models
- Frontier LLMs may be post-trained on protein data: SOD1 ranking test matches ESM2 — alex_peys · 2026-11-15
- Ernie 5.5 Flash Preview Spotted on RuneBench, Reportedly Near GLM 5.2 Level — mariofilhoml · 2026-11-15
- GLM-5.3 on Two DGX Stations Peaks at 607.8 tok/s but 32 Concurrency Causes 453s Latency — petrusenko_max · 2026-11-15
- Unverified leak claims Gemini 4 Hydrogen targets fast agentic execution — feblet · 2026-11-15
- Gemini 4 Carbon takes the MathArena crown but costs 15 times more than GPT-6.1 Sol — The Decoder · 2026-11-15
- Anthropic says outdated prompting habits are slowing down Claude Opus 5.5 — The Decoder · 2026-11-15