New Open Models: TwIL-8B and Ripple-65B Preview

Two notable open model releases have entered the AI ecosystem, offering developers new options for reasoning and dense model architectures. According to MarkTechPost, webAI and Ricursive AI both introduced significant models designed to push performance boundaries in specific domains.

What happened

webAI released TwIL-8B, an 8.8-billion parameter reasoning model built by post-training IBM Granite 4.2 8B. According to MarkTechPost, the model achieved a score of 98.2 on MATH-500 and 81.6 on LiveCodeBench. However, the model lags behind other models of a similar size when evaluated on knowledge benchmarks.

In a separate release covered by MarkTechPost, Ricursive Intelligence launched the Ripple-65B Preview. This offering is a 65-billion dense open-weight large language model released under the Apache 2.0 license. It managed a score of 80.5 on Terminal-Bench 2.1.

Why it matters

The introduction of these models highlights ongoing advancements in specialized reasoning and open-weight architectures. webAI's TwIL-8B demonstrates strong capabilities in mathematical and coding tasks despite its compact size. Meanwhile, Ricursive AI's Ripple-65B Preview outperforms several comparable US open models up to 15 times its size on specific terminal benchmarks, highlighting efficiency gains in dense model designs.

What it means for developers and businesses

For developers and small-business owners, these releases expand the toolkit for local and open-source deployments. TwIL-8B offers strong mathematical and coding reasoning at an 8.8B scale, which can be easier to host than massive frontier models. Ripple-65B provides an Apache 2.0 licensed dense option that performs competitively on command-line and technical benchmarks, making it a compelling candidate for engineering workflows.

What to watch next

Developers should monitor how these models perform in real-world application testing beyond standardized benchmarks. Future updates will likely address knowledge gaps in models like TwIL-8B and bring Ripple-65B closer to a full production release as feedback from the preview accumulates.