whatmodel.sh / Ripple-65B Preview

Ripple-65B Preview is an open-weights language model from Ricursive Intelligence, released November 2026. Its published size is 65B parameters and a 1M-token context window. Licence: apache-2.0.

Catalogue identity notes

BF16 instruction-tuned checkpoint (this repo, created 2026-11-04, 129.6 GB). Ricursive also publishes Ripple-65B-Preview-NVFP4 ('Experimental'; 'NVFP4 for the MLP layers while keeping attention in BF16'; 57.2 GB; its card gives no serving command), the pretrain-only Ripple-65B-Base-Preview (not added) and a DFlash2 draft. Text-only. Card: 'Our Preview release has gone through relatively light posttraining, with a focus on agentic coding.' Launch blog (ricursive.com/ripple): 'available on Hugging Face, Together AI, GMI Cloud, OpenRouter, and Fireworks' - no first-party API and the host pages were not checked, so api is null. The BF16 repo's card has no vLLM/SGLang command (only custom inference/server code), so local routes are the official Ollama library tags.

Model information sources

Release date ↗ · Weights licence ↗ · Parameter count ↗

Dense. Card table: Total params 65B, Active params 65B. HF safetensors metadata at b172323: 64,782,689,360 BF16 parameters.

44.0±7.2#675 of 974 LLMsfrom 1 benchmark result across 1 source* thinly tested, treat the rank as provisional

Score within the LMArena (text) comparison set only.

$ how it compares

The LLMs ranked either side of Ripple-65B Preview, out of 1206 with a score.

#modelscorewhereRAMreleased
848Infinity-Instruct-3M-0625-Yi-1.5-9B BAAI44±2local~7.2GB2024-07
849LFM2 2.6B Liquid AI44±1local~2.1GB2025-09
850Ripple-65B Preview Ricursive Intelligence44±7local~42GB2026-11
851CodeGemma 7B IT (v1.0) Google44±1local~7.1GB2024-03
852DeepSeek-Coder 1.3B Instruct DeepSeek44±1local~2.9GB2023-10

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasksnot ranked on this task yet · 6 tests
testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab76.5% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 114/140hard · avg 19%—not counted yet: no other model in the same run to compare it with
Terminal-Bench 2.1 lab80.5% ✎▁·▁·▁▁▄▃▅█▆█ beats 46/68very hard · avg 7.5%—not counted yet: no other model in the same run to compare it with
SWE-bench Pro (split unspecified) lab61.4% ✎▁·▁▁▁▁▁▄█▁▁▁ beats 48/64very hard · avg 4.0%—not counted yet: no other model in the same run to compare it with
MCP-Atlas (public set) lab71.7% ✎▂·▁▂▂·▁▃█▆▆▃ beats 27/41hard · avg 31%—not counted yet: no other model in the same run to compare it with
DeepSWE v1.1 (resolved) lab44.3% ✎▂····▃▁▁▄█▆█ beats 4/36—not counted yet: no other model in the same run to compare it with
τ³-bench Banking lab31.3% ✎▂▂·▅█▂▃▃▃▆▂▆ beats 23/25very hard · avg 5.4%—not counted yet: no other model in the same run to compare it with
codingnot ranked on this task yet · 4 tests
testresultvs the others in its rundifficultycountswhy
SWE-bench Verified lab76.5% ✎▁▁▁▁▁▁▂▄█▆▁▁ beats 114/140hard · avg 19%—not counted yet: no other model in the same run to compare it with
SWE-bench Pro (split unspecified) lab61.4% ✎▁·▁▁▁▁▁▄█▁▁▁ beats 48/64very hard · avg 4.0%—not counted yet: no other model in the same run to compare it with
SciCode (level unspecified) lab39.4% ✎▁····▃▃▄▄█▆▃ beats 27/42very hard · avg 13%—not counted yet: no other model in the same run to compare it with
DeepSWE v1.1 (resolved) lab44.3% ✎▂····▃▁▁▄█▆█ beats 4/36—not counted yet: no other model in the same run to compare it with
mathnot ranked on this task yet · 4 tests
testresultvs the others in its rundifficultycountswhy
AIME 2025 lab98.3% ✎▁·▁▁▁▃▅▇█▆▃▃ beats 134/134medium · avg 43%—not counted yet: no other model in the same run to compare it with
AIME 2024 lab97.1% ✎▁▂·▂▁▂▂▄█▅▆▃ beats 86/86medium · avg 59%—not counted yet: no other model in the same run to compare it with
AIME 2026 lab95.0% ✎▁▁▁▂▂▂▆█▇▂·▁ beats 40/54medium · avg 49%—not counted yet: no other model in the same run to compare it with
HMMT February 2026 lab90.5% ✎▃▃▁▄▇▇▇█▅▃▃▃ beats 32/38hard · avg 31%—not counted yet: no other model in the same run to compare it with
other tests6 not in a task · 77% of its score
testresultvs the others in its rundifficultycountswhy
Icelandic LLM Leaderboard average (Miðeind, Sept 2026 suite)7.7% ✎▃▂▄▂▃▃▂▇▇█▇▁ beats 1/47very hard · avg 14%▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 77%typical
GPQA Diamond lab85.2% ✎▁▁▁▁▂▃▅▆▇▇█▂ beats 235/336hard · avg 39%—not counted yet: no other model in the same run to compare it with
MMLU lab89.6% ✎▁▁▁▁▁▂▂▅██▆▄ beats 205/215medium · avg 73%—not counted yet: no other model in the same run to compare it with
MMLU-Pro lab84.2% ✎▁▁·▁▁▁▃▂▄▅█▁ beats 163/189medium · avg 55%—not counted yet: no other model in the same run to compare it with
IFBench (aggregation unspecified) lab79.2% ✎▂▆▂▃▄▄▂▃▄▆█▇ beats 79/88medium · avg 44%—not counted yet: no other model in the same run to compare it with
AutomationBench (Public) lab36.4% ✎█····▄··▄·▄▄ beats 3/5—not counted yet: no other model in the same run to compare it with

$ where to run it

Memory is what a version takes in GPU memory (or a Mac's unified memory) with an 8k-token conversation. With less, it runs partly from system RAM, more slowly. Enter your machine to see what runs at full speed.

Ollama · recommended

version

  1. Install Ollama

    download it (macOS / Windows) ↗

    Linux
    curl -fsSL https://ollama.com/install.sh | sh
  2. Run Ripple-65B Preview · needs 42.2 GB memory, 38.8 GB download · from ollama.com
    chat in the terminal
    ollama run ripple-preview:65b-q4_K_M

Model repository and documentation: huggingface.co/ricursiveAI ↗

$ how the score was worked out · Miðeind Icelandic LLM Leaderb… 77%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
Miðeind Icelandic LLM Leaderboard1CC BY 4.0
prior · pull toward the middle
weighted average= 44.0

not counted (16)

ricursiveAI/Ripple-65B-Preview model cardMMLU [correct]89.6%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardMMLU-Pro [correct]84.2%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardGPQA Diamond [correct]85.2%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardAIME 2024 [correct]97.1%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardAIME 2025 [correct]98.3%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardAIME 2026 [correct]95.0%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardHMMT February 2026 [correct]90.5%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardTerminal-Bench 2.1 [correct]80.5%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardSWE-bench Verified [correct]76.5%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardSWE-bench Pro (split unspecified) [correct]61.4%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardDeepSWE v1.1 (resolved) [correct]44.3%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardSciCode (level unspecified) [correct]39.4%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardτ³-bench Banking [correct]31.3%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardAutomationBench (Public) [correct]36.4%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardMCP-Atlas (public set) [correct]71.7%fewer than three models have this test yet, or this source ran it for this model alone
ricursiveAI/Ripple-65B-Preview model cardIFBench (aggregation unspecified) [correct]79.2%fewer than three models have this test yet, or this source ran it for this model alone

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json