TwIL-LM3
A 3-billion-parameter small language model specialized in formal reasoning, released by webAI. After Q4_K_M quantization it occupies only 1.78 GiB, enabling deployment on 4 GB of VRAM or CPU-only hardware. It focuses on structured tasks including formal reasoning, rule induction, and semantic parsing.
Timeline (last 90 days)
webAI has launched the TwIL-LM3 model with a parameter scale of 3 billion (3B).
On the Rule Induction task, TwIL-LM3 scored 96.4 while gpt-oss-120b scored 65.2.
In terms of throughput, TwIL-LM3 reaches 32.9 answers/sec, while gpt-oss-120b achieves 12.6 answers/sec.
On broader comprehensive benchmark aggregate scores, the 120B model still leads TwIL-LM3.
TwIL-LM3's wins are strictly limited to the two dimensions of narrow formal reasoning tasks and operational efficiency.
For application scenarios such as narrow formal reasoning, compliance rules, contract logic, and research validation, a specialist model that can run locally is superior to a massive general-purpose model hidden behind an API.
The TwIL-LM3 model is available for download on Hugging Face at huggingface.co/webAI-Official/TwIL-LM3.
All Facts (7)
On broader comprehensive benchmark aggregate scores, the 120B model still leads TwIL-LM3.
50%UnverifiedOn the Rule Induction task, TwIL-LM3 scored 96.4 while gpt-oss-120b scored 65.2.
50%UnverifiedIn terms of throughput, TwIL-LM3 reaches 32.9 answers/sec, while gpt-oss-120b achieves 12.6 answers/sec.
50%UnverifiedThe TwIL-LM3 model is available for download on Hugging Face at huggingface.co/webAI-Official/TwIL-LM3.
50%UnverifiedTwIL-LM3's wins are strictly limited to the two dimensions of narrow formal reasoning tasks and operational efficiency.
50%UnverifiedFor application scenarios such as narrow formal reasoning, compliance rules, contract logic, and research validation, a specialist model that can run locally is superior to a massive general-purpose model hidden behind an API.
50%UnverifiedwebAI has launched the TwIL-LM3 model with a parameter scale of 3 billion (3B).
50%