ricursiveAI/Ripple-65B-Preview
- Runs here
- 6of 6
- List price of its runs
- $0.025
- Total time
- 4 min 19 s
- Price
- $0.60/M out
- Hacker News
- 343pts · 3 stories
Attempts
Pelican riding a bicycle (SVG)
2 min 14 s$0.01017k tokensAPI
Wristwatch that shows the current time
29 s$0.0046.2k tokensAPI
Landing page for a cargo bike workshop
78 s$0.00914k tokensAPI
14 s$0.0012.3k tokensAPI 
3 s<$0.001490 tokensAPI Patchbay is three people and a bench in Rotterdam. We repair vintage synthesizers of all kinds, from a Moog that needs one capacitor to a modular system that hasn't been run in thirty years. We started because older instruments deserve to keep working. The circuitry is simple, but the failures are personal: a dead pot in a favourite key, a CV rail that drifts, a ribbon controller that was saturated in smoke. We treat each case the way our own instruments would need to be treated. A workshop is a good place for slow work and long conversations. We're always glad to open a system on the bench, find out what it actually needs, and explain why. If you have something unwell, send it over or come visit us in Rotterdam.
2 s<$0.001156 tokensAPI
About
Ricursive says Ripple-65B Preview is a general-purpose version of the model behind its chip design platform, aimed at agentic coding. Ricursive says it scored 80.52 on TerminalBench 2.1, ahead of larger US open-weight models such as Reflection Beam 501B (80.1). It is a dense 65B model with a context of 1,048,576 tokens, listed on OpenRouter at $0.2 in and $0.6 out per million tokens.
The top Hacker News story reached 340 points. User thesnarkitecht, who has high-memory Mac Studios, wrote "it is great that there is finally a model that will run at usable speeds with high quality". On X, LLMJunky ran NVFP4 on a single RTX 6000 and wrote "I just cannot use models this slow." pratikg measured a steady 7 tok/s with a 4-bit MLX build on a 64GB M5 Pro Mac mini.
All six of our runs worked, in 4 min 19 s and $0.025 at list price. The pelican took 2 min 14 s and 17k output tokens, against a median of 70 s and 9.9k over 47 models. Breakout took 14 s and 2.3k tokens, against a median of 2 min 29 s and 21k. The watch page has two grey ovals beside the dial, and the landing page headline runs off the right edge.
Sources: OpenRouter · Launch post · HN thread · @LLMJunky on X. Written by Claude Sonnet 5.5 on 8 Nov 2026 from these sources and our runs, then checked by Claude Opus 5.5 and by a script that matches every number and quote to a source.
Specs
- Price in
- $0.20 / M
- Price out
- $0.60 / M
- Context
- 1,048,576 tokens
- Max output
- 131,072 tokens
- Input
- text
- Made in
- —
- Served in
- US · 3 providers
- OpenRouter
- ricursiveai/ripple-65b-preview
On Hacker News
343 HN pts · 3 stories
Top story: Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding (341 points)
On Hugging Face: 495 likes, 7k downloads in 30 days
Compare
Also side by side with Particle-1, GPT-6.1 Terra, Kimi K3.1 Pro, DeepSeek V4.1 Flash 1017, Muse Spark Open, Gemini 4 Argon, LFM2.5-24B-A2B, Claude Haiku 5.5, Mistral Large 4, GPT-6.1 Sol, Claude Opus 5.5, MiMo v2.6 Pro, Gemini 3.8 Flash, Claude Fable 5.1, GLM 5.3 Flash, GLM 5.3 and Kimi K3.