Release119 comments in the community
Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Hacker News: Front Page · By Philpax ·
The Brief
- Ricursive released Ripple-65B Preview, an open-weight model that its maker describes as state of the art on agentic coding within a narrow U.S.-made open-weight category.
- The model is available on Hugging Face under ricursiveAI/Ripple-65B-Preview, and a GGUF build has been posted by AtomicChat.
- Commenters report it is a 65B dense model and cite a Terminal Bench 2.1 score of 80.5, but the benchmark comparison omits several widely used open models.
Why it matters
Developers who run open-weight models locally or in air-gapped settings get another large option, but the claimed lead rests on a narrow benchmark selection, so its real standing should be checked against independent evaluations.
Who’s affected
- Developers running open-weight models locally
- Teams that need U.S.-origin open-weight models
- Apple Silicon users with high-memory machines
- Ripple-65B-Preview
- #open-weights
- #llm
- #agentic-coding
- #local-inference
- #benchmarks
What developers are saying
Based on 119 comments · Mixed reactions
Commenters welcome another competitive open-weight model, especially one that could run on high-memory Macs. Many criticise the benchmark selection for omitting Qwen and other leading open models, and some question what the U.S. qualifier means and whether the SOTA claim is meaningful. Several note that a 65B dense model will be slow on consumer hardware.
- Narrow benchmark comparison35%
- Welcome for new open model25%
- Local hardware fit20%
- Meaning of U.S. and SOTA15%
- Outdated benchmarks10%
In favour
- + Another competitive open-weight model in the field
- + Roughly 200GB-or-less footprint suits high-memory Mac setups
Concerns
- − Comparison omits Qwen and other leading open models, looking cherry-picked
- − Terminal Bench 2.1 is considered outdated relative to newer versions
- − A 65B dense model will be slow on consumer hardware
- − The U.S. qualifier and narrow SOTA category are seen as diluting the claim
“Congratulations I guess but it's a pretty bad look to not even mention those other open weight models.”
“Congratulations to the recursive team for building an open model that will actually run!”
“When a post doesn’t mention Qwen for comparison, you know the performance isn’t going to be great.”