Reported News
Ripple-65B Preview: a dense 65B open-weight model that reports 80.5 on Terminal-Bench 2.1, Apache 2.0
Ricursive Intelligence released Ripple-65B Preview on November 4, 2026, a dense 65B-parameter open-weight model (Apache 2.0) that it says scores "80.52" on TerminalBench 2.1, ahead of much larger US open-weight models. It supports up to 1M tokens of context and is on Hugging Face, Together AI, GMI Cloud and OpenRouter.
- Changes reported
- 6
- listed below, each from a source we read
- We tested
- Not tested yet
- reported from sources; we did not run it ourselves
- Gaps we flag
- 4
- things we could not verify
- Sources
- 3
- 2 primary, linked below
What changed
| # | What the source says | Status |
|---|---|---|
| 01 | New model from a new lab: a 65B-parameter dense Transformer (not Mixture-of-Experts) pretrained from scratch on 12 trillion tokens of mostly English text and code, with support for up to 1M tokens of context. | Reported |
| 02 | Ricursive reports 80.52 on Terminal-Bench 2.1, above the self-reported scores it lists for Reflection Beam 501B (80.1), Muse Spark Open 816B (76.4), Inkling 975B (63.8) and Nemotron 3 Ultra 550B (56.4). | Reported |
| 03 | The model card also lists 76.5 on SWE-bench Verified and 61.42 on SWE-bench Pro v1.0, both below Reflection Beam's 81.6 and 66.1 in the same table, so the coding lead is benchmark-specific. | Reported |
| 04 | Weights are Apache 2.0 on Hugging Face: a BF16 instruction-tuned model, an NVFP4 checkpoint and a BF16 base model. Ricursive says it can run locally on an NVIDIA DGX Spark. | Reported |
| 05 | Hosted access on day one through Together AI, GMI Cloud and OpenRouter, with Fireworks listed as going live later the same day. | Reported |
| 06 | Ricursive puts its full R&D cost at 32 million dollars and says Terminal-Bench 2.1 was reached with 1.4M RL rollouts; the post-training is described as relatively light and focused on agentic coding. | Reported |
reported by the sourcemeasured or run by usnot verified
Primary source: Ricursive Intelligence: Introducing Ripple-65B Preview. We are the fast, independent index; the vendor’s own page is the source of record for this release.
In the source’s words
“The model is pretrained on 12 trillion tokens with support for up to 1M context length.”
“Our Preview release has gone through relatively light posttraining, with a focus on agentic coding.”
What we have not verified
- We did not run Ripple-65B Preview; every benchmark score is Ricursive's own, and the model card says some comparison scores come from its in-house eval flow rather than official model cards.
- The announcement page has no machine-readable publish time; the date is from the page's November 4, 2026 release and the Hugging Face repository.
- We did not check hosted pricing or rate limits on Together AI, GMI Cloud, OpenRouter or Fireworks.
- The 32 million dollar cost and the one-tenth comparison are Ricursive's estimates, not audited figures.
This brief is Reported: written from the sources below, not from running the release ourselves.
Sources
- Ricursive Intelligence: Introducing Ripple-65B Previewprimary source · fetched 2026-11-04
- Hugging Face model card: ricursiveAI/Ripple-65B-Previewprimary source · fetched 2026-11-04
- Hacker News discussion (signal)source · fetched 2026-11-04
Related
Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/ripple-65b-preview-open-weight-coding-model.md
Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.
Newsletter
The Weekly Verdict. Free, every Monday.
A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.
Free. Double opt-in. One click to unsubscribe.