cd /news/artificial-intelligence/webai-open-sources-twil-lm3-a-1-7b-f… · home topics artificial-intelligence article
[ARTICLE · art-90921] src=cryptobriefing.com ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

WebAI open-sources TwiL-LM3, a 1.7B formal reasoning model that outperforms OpenAI’s gpt-oss-120B

WebAI, an Austin-based AI company, released its TwiL-LM family of formal-logic reasoning models on August 10, with the flagship TwiL-LM3, a 3B-parameter model, outperforming OpenAI's gpt-oss-120B on four out of five benchmarks, scoring 96.4 accuracy on rule induction tasks versus 65.2 for the OpenAI model, while processing 32.9 answers per second compared to 12.6. The models, available in 1.7B and 3B sizes on Hugging Face in GGUF formats under a non-commercial license, are designed for on-device inference and formal reasoning tasks like compliance checks and contract analysis.

read2 min views1 publishedAug 10, 2026
WebAI open-sources TwiL-LM3, a 1.7B formal reasoning model that outperforms OpenAI’s gpt-oss-120B
Image: Cryptobriefing (auto-discovered)

Via webai.com

The Austin-based AI company claims its 3B-parameter model beats a rival 40 times its size on four out of five formal reasoning benchmarks, all while running on consumer hardware.

A model with 3 billion parameters just embarrassed one with 120 billion. WebAI, an Austin-based AI company, released its TwiL-LM family of formal-logic reasoning models on August 10, and the benchmarks tell a story that should make anyone paying cloud compute bills a little uncomfortable.

The flagship TwiL-LM3, a 3B-parameter model, outperformed OpenAI’s gpt-oss-120B on four out of five benchmarks in webAI’s formal-reasoning suite. That’s a model roughly 40 times smaller beating a giant on nearly every test thrown at it. On rule induction tasks specifically, TwiL-LM3 scored 96.4 accuracy compared to 65.2 for the OpenAI model.

Small model, big throughput #

Raw accuracy is only half the picture. TwiL-LM3 also processed roughly 32.9 answers per second, compared to approximately 12.6 for OpenAI’s model. That’s about 2.6 times faster throughput from a model that can actually run on hardware you already own.

The TwiL-LM family ships in two sizes. The 1.7B-parameter variant is optimized for phones, targeting on-device inference for mobile applications. The larger 3B TwiL-LM3 is the benchmark-topping version, fine-tuned using SmolLM3-3B as a base model and enhanced with LoRA adapters. LoRA, short for Low-Rank Adaptation, is a technique that lets developers fine-tune large models efficiently by only training a small number of additional parameters rather than retraining the entire network.

Both models are available on Hugging Face in GGUF formats, which are designed for local deployment. The license is non-commercial, so hobbyists and researchers can experiment freely, though enterprise adoption would require a separate arrangement.

Why formal reasoning matters now #

WebAI isn’t trying to build a chatbot that writes poetry or generates marketing copy. The TwiL-LM models are specifically tailored for deductive reasoning tasks: compliance checks, contract analysis, and any workflow where logic needs to be airtight rather than creative.

CEO David Stout highlighted the models’ utility for writing and tool-calling applications, while CIO Dr. Paul J. Maykish emphasized the strategic importance of narrow, specialized models over bloated generalist systems.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our

Editorial Policy.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @webai 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/webai-open-sources-t…] indexed:0 read:2min 2026-08-10 ·