Chip company releases open AI model that scores high on coding test

Ricursive Intelligence, which designs AI chips for a living, has trained and released an open language model called Ripple-65B Preview. According to Runtime Wire, the release took place on November 4, 2026.

The model is available on the developer site Hugging Face under the Apache 2.0 license. That means others can freely download it, use it and build on it.

What kind of model is it?

Ripple is a so-called dense Transformer with 65 billion parameters. Parameters are the settings an AI model learns during training, and they are often used as a rough measure of a model's size. Dense here means the model is not built as a mixture-of-experts, where only part of the model is in use at a time. According to The Decoder, it was pretrained on 12 trillion tokens (small pieces of text) and optimized for Nvidia hardware.

Runtime Wire writes that the model is derived from a proprietary model for chip design.

High score on coding test

Ricursive says the model scored 80.52 on TerminalBench 2.1, a test of how well AI can solve coding tasks. The company calls it the best result among open American models. The result was achieved with Terminus 2 and OpenCode as the harness, the test framework that controls how the model runs the tasks.

According to Runtime Wire, Ricursive also claims that Ripple beats competitors with 975 billion parameters on the same test.

Caveats

Both the benchmark result and the cost figure come from Ricursive itself. Runtime Wire stresses that independent runs are needed to confirm the TerminalBench result.

On other benchmarks, according to The Decoder, Ripple cannot match the likewise newly released Inkling 2 Small. The sources do not make clear which tests are involved or how large the gap is. The Decoder assesses that Ripple is probably easier to run on users' own hardware.

Ricursive also says that total development and training costs were $32 million, and that the amount covers all research and development expenses.

What does it mean for you?

For ordinary users, the release is unlikely to change anything in the short term. But an open model with a free license gives companies and developers the option to run and customize a coding model themselves, instead of only renting access from large providers.