First screen, captured 9 days ago. Open the video.

Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding

What HN thinks

State of the art, terms and conditions apply, see footnote 7 regarding nationality.

SOTA claim is overly narrowed
Many mocked the 'U.S. open-weight' qualifier as category-shrinking that drains meaning from SOTA and 'frontier', and faulted the cherry-picked comparisons that omit Qwen and other leading open models.
US origin matters to some buyers
Government and regulated users say they are barred from Chinese models even when self-hosted, so a US-made Apache-2.0 model fills a real niche; others argue origin is irrelevant for weights you run yourself.
Dense 65B is awkward locally
Hobbyists note a dense 65B model will be slow on consumer hardware and Macs compared with MoE models like Qwen3.8 Flash Next, though quantisation and MTP may help.

Consensus: Most agree the benchmark framing is cherry-picked and that real comparisons against Qwen and other top open models are needed before judging it, while still welcoming another open-weight entrant.

Recommended in the comments

Qwen3.8-27B (product): Cited as the obvious comparison point that runs well locally with strong quality.
llama.cpp ↗ (tool): Mentioned for fast local inference and new expert-caching features for MoE models.

Summarised by Claude from the comment thread 8 days ago.

Points: 327
Comments: 116
Best rank: #1
Posted: 2026-11-04 by Philpax
Category: AI & ML

ripple-65b open-weight-models llm agentic-coding terminal-bench reinforcement-learning ricursive-intelligence

All captures:

Full page

Captured 9 days ago. Click for the whole screenshot.

Looks similar

Screenshot of ps-2.kev009.com: OS/2 Warp, PowerPC Edition

OS/2 Warp, PowerPC Edition (ps-2.kev009.com)

111 points by Lammy 3 years ago | 53 comments

Screenshot of blog.vllm.ai: vLLM large scale serving: DeepSeek 2.2k tok/s/h200 with wide-ep

vLLM large scale serving: DeepSeek 2.2k tok/s/h200 with wide-ep (blog.vllm.ai)

138 points by robertnishihara 10 months ago | 46 comments

Screenshot of research.meta.ai: Muse Spark Open

Muse Spark Open (research.meta.ai)

884 points by bigwheels 29 days ago | 446 comments

Meta research blog post announcing Muse Spark Open, an open-weight large language model Meta calls the most capable US open-weight model to date.

muse-spark-open meta open-weight-models llm

Screenshot of openssh.org: OpenSSH 10.5/10.5p1

OpenSSH 10.5/10.5p1 (openssh.org)

103 points by voxadam 3 months ago | 32 comments

Screenshot of research.meta.ai: Muse Spark Open, the Most Capable US Open‑Weight Model to Date

Muse Spark Open, the Most Capable US Open‑Weight Model to Date (research.meta.ai)

38 points by meetpateltech 29 days ago | 1 comments

Meta blog post announcing Muse Spark Open, an open-weight AI model billed as the most capable US open-weight model to date.

muse-spark meta open-weight-models llm

Screenshot of seangoedecke.com: OpenAI's new open-source model is basically Phi-5

OpenAI's new open-source model is basically Phi-5 (seangoedecke.com)

372 points by emschwartz 1 year ago | 200 comments

Sean Goedecke's blog post arguing that OpenAI's open-weight gpt-oss models resemble Microsoft's Phi series, strong on benchmarks but weak on out-of-domain knowledge.

gpt-oss openai phi open-weight-models