webAI Open-Sources TwiL-LM: 3B On-Device Model Out-Reasons 120B
rohanpaul_ai · x · 2026-08-11
webAI has open-sourced TwiL-LM, a family of formal-logic models (1.7B and 3B parameters) designed for deductive reasoning. The 3B model outperforms gpt-oss-120b—which is 40× larger—on four out of five formal-reasoning benchmarks while delivering 2.6× faster inference.
- Core Capability: Translates plain English into formal logic to verify multi-step deductive conclusions, powering reliable tool calling, compliance checks, and contract logic.
- On-Device: The 1.7B variant is roughly 1GB and runs entirely on consumer hardware like iPhones, enabling private, edge-based enterprise deployments.
- Training Pipeline: Fine-tuned using a proprietary data engine with verified datasets rather than scraped internet data, utilizing a 72M parameter LoRA adapter.
More from Models
- Web Design Benchmark: Muse Glimmer 30B vs. Qwen 3.6 27b vs. DeepSeek — ShadyShroomz · 2026-08-11
- Muse Glimmer Local Coding Test: Runs on 20GB RAM, Lags Behind Qwen — curiousily_ · 2026-08-11
- Perplexity Agent API Integrates Kimi K3 with Automated Data Story Workflow — AravSrinivas · 2026-08-11
- Renaming a PDF Boosts LLM Scores: A Weird Flaw in AI Evaluation — generativist · 2026-08-11
- DeepSeek Open-Sources DeepSeek-OCR 2: Visual Causal Flow for Markdown Conversion — tom_doerr · 2026-08-11
- Prediction: Anthropic Will Split Chat and Agentic Models — natanielruizg · 2026-08-11