webAI Open-Sources TwIL-8B, a Small Model Approaching Frontier Coding and Math

1 reports archived1 independent sourcesupdated 11/6/2026, 03:25:09

Synthesis & Latest Updates

1 Sources Cross-Validated

webAI has open-sourced TwIL-8B, an 8.8B-parameter model built on IBM's Granite 4.2 8B and tuned for math, coding and instruction following. It reportedly fits in a 4.98 GiB quantized file that runs on CPUs or modest GPUs, scoring 81.6% on LiveCodeBench and reaching roughly 92-93% of frontier-model performance on coding and instruction following.

LATEST/webAI has open-sourced TwIL-8B, an 8.8B-parameter model built on IBM's Granite 4.2 8B and tuned for math, coding and instruction following. It reportedly fits in a 4.98 GiB quantized file that runs on CPUs or modest GPUs, scoring 81.6% on LiveCodeBench and reaching roughly 92-93% of frontier-model performance on coding and instruction following.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

Rohan Paul (@rohanpaul_ai)T2Nov 6, 03:25·78 pts
webAI Open-Sources TwIL-8B, a Small Model Approaching Frontier Coding and Math
- The 8.8B TwIL-8B fits in a 4.98 GiB quantized file and runs on CPUs or modest GPUs
- It scores 81.6% on LiveCodeBench, beating Qwen3.5-9B and the 3.5x larger Gemma 4 31B
- A training stage rewarding shorter reasoning cuts AIME 2026 token usage by 38% versus Qwen3.5-9B
View Related Reports & Analysis Original Source