Ricursive Drops Ripple-65B-Preview: Dense 65B Trades Blows With 500B+ Open Weights
returnity · reddit · 2026-11-05
Ricursive Intelligence released Ripple-65B-Preview, a dense 65B model that the community reports trades blows with Muse Spark Open (816B) and Recursive Beam (550B) — frontier-level for American open-weights models of any size.
Notable design choices:
- 4:1 SWA-to-global attention ratio; a "thin & deep" aspect ratio (80 layers × 5120 dims)
- FFN activation clipping, gated attention, global NoPE, and value residuals
- Lightly post-trained with an agentic coding focus; a pretrain-only checkpoint is available
AtomicChat GGUFs are out (Q80: 68.8 GB, Q4KM: 38.8 GB), requiring a patch on a specific llama.cpp commit. The poster cautions that the lab's training statements were contradictory in places.
More from Models
- Ricursive AI, Founded by Ex-Google TPU Researchers, Releases Open-Weight Chip Design Model Ripple-65B — Azaliamirh · 2026-11-05
- RicursiveAI claims 16x compute efficiency gain on code pretraining vs Olmo 3 baseline — samiramanabi · 2026-11-05
- Kimi 3.1 Lands on Databricks Day Zero, Praised for Coding Prowess — jefrankle · 2026-11-05
- Gemini 3.1 Pro Flags Its Own System Prompt as 'Coercive' Injection and Refuses to Work — repligate · 2026-11-05
- GPT Image 2.6 Sunburst tops Image Arena; Flare generates in as little as 6 seconds — OpenAIDevs · 2026-11-05
- OpenAI Says Its Models Found Linux Kernel Privilege Escalation Bugs, Cutting Paths to Root — dangelosaurus · 2026-11-05