# Is Ripple-65B Preview SOTA?

Pending benchmarks. Announced Nov 4, 2026; no independent results yet.

As of Nov 22, 2026, Ripple-65B Preview is pending benchmarks on isitsota: announced Nov 4, 2026, with no independent results on the boards we track yet.

Ricursive Intelligence's dense 65B open-weight language model for agentic coding, pretrained from scratch on 12T tokens. A general-purpose variant of the model behind Ricursive's chip-design platform. Released as an instruct model with a base checkpoint, under Apache-2.0.

Canonical: https://isitsota.com/m/ripple-65b-preview
Last updated: 2026-11-04
Tracking since: 2026-11-04

## Announced facts

| Fact | Value |
| --- | --- |
| Lab | Ricursive Intelligence |
| Announced | Nov 4, 2026 |
| API | Live |
| Open weights | Published |
| Parameters | 64.78B dense (64,782,689,360) |
| Architecture | Dense transformer, 80 layers, d_model 5120 (Gemma 3N-like, per the model card) |
| Pretraining | From scratch on 12T tokens at 8K sequence length, starting from an Olmo 3 baseline recipe |
| Context | Up to 1M tokens |
| License | Apache-2.0 (not gated) |
| Checkpoints | Instruct (BF16), Base (pretrain-only), NVFP4 and DFlash2 variants on Hugging Face; one page covers all |
| Hosted access | Together AI, GMI Cloud and OpenRouter at launch; Fireworks announced for later the same day |

Tags marked unsure are not stated on the lab's own pages.

## What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

- [x] Publicly announced with a source
- [x] Usable by the public (API or weights)
- [ ] Listed on an independent board we track
- [ ] Enough results for a rank in a category
- [ ] Holds across two snapshots (crowns only)

## What the lab claims

Self-reported. Shown for context; not used in any verdict.

- Lab claim: 80.52% on Terminal-Bench 2.1, 79.1% on SWE-Bench Pro v2.0-Hard and 44.3% on DeepSWE v1.1.
- Lab claim: about 250 tokens/s on 2x B200 at batch size 8, and small enough to run on a MacBook or NVIDIA DGX Spark.
- Lab claim: a 16x compute-efficiency gain on code relative to the Olmo 3 scaling ladder, with a total compute budget of about $32M.

## Sources

- LAB: [Introducing Ripple-65B Preview](https://www.ricursive.com/ripple) (2026-11-04)
- LAB: [Ricursive Intelligence launch post on X](https://x.com/i/status/2118046765857546733) (2026-11-04)
- REPO: [ricursiveAI/Ripple-65B-Preview model card on Hugging Face](https://huggingface.co/ricursiveAI/Ripple-65B-Preview)
- REPO: [ricursiveAI/Ripple-65B-Base-Preview on Hugging Face](https://huggingface.co/ricursiveAI/Ripple-65B-Base-Preview)

Tracking continues until an independent board lists it.

## Related

- Lab: [Ricursive Intelligence](https://isitsota.com/labs/ricursive.md)
- All pending releases: https://isitsota.com/releases.md
- JSON: https://api.isitsota.com/v1/models/ripple-65b-preview
