Ricursive (AlphaChip creators) releases Ripple-65B Preview, an Apache-2.0 dense 65B model trained from scratch for $32M
Partly confirmed
The takeaway
Its own benchmarks claim the US open-weights coding lead; the company admits difficulty reproducing agent benchmarks.
Status
- Claim
Partly confirmed
- Our reporting
- Medium confidence
- Importance
- 3 of 5
- Last verified
- 5 November 2026
Your AI and this story
- GPT-6 Astra188 days after its cutoff
- Claude Opus 5.5127 days after its cutoff
- Gemini 4 Argon218 days after its cutoff
- Grok 4.7157 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 127 days before it.
Key facts
- Dense, 80 layers, 12T training tokens; advertised 1M context (OpenRouter caps at 16,384 tokens)
- Vendor numbers: Terminal-Bench 2.1 80.5, SWE-bench Pro v2.0 Hard 79.1, DeepSWE 1.1 44.3 (Muse Spark Open 60.0), HMMT Feb 2026 90.5
- Model card gives different rival scores from the launch graphic (VentureBeat)
- OpenRouter (GMI Cloud): $0.20 input / $0.60 output / $0.05 cached per 1M tokens; weights at huggingface.co/ricursiveAI/Ripple-65B-Preview
What happened
A rare US-made, permissively licensed open model trained from scratch by a startup, at a disclosed cost. Benchmarks are vendor-only so far; it is not on Artificial Analysis’s open-weights list yet.
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 1 primary, 1 press
Primary
- Hugging Face: ricursiveAI/Ripple-65B-Previewhuggingface.co, code
Press
- VentureBeat: Ripple-65B Preview claims US open coding leadventurebeat.com, press
Changes
- Filed