Leaderboard / Ricursive Intelligence

Ripple 65B Preview

Ricursive Intelligence · Ripple · released 2026-11-04 · data checked 2026-11-13

Ripple 65B Preview is a 64.8B-parameter dense model from Ricursive Intelligence, released 2026-11-04 under the Apache-2.0 license. It scores 74.6 on the Small LM Index, #1 of 9 large models. It needs about 41.9 GB of memory at 4-bit with an 8K-token context.

Dense model trained from scratch (12T tokens) by a five-person team, starting from an Olmo 3 recipe. Preview release with light post-training focused on agentic coding; scores at maximum effort. A base model is also published.

Small LM Index
74.6
#1 of 9 scored large models
Likely 73.9–76.6 · from 4 of 6 core benchmarks
Parameters
64.8B
Dense
Price per 1M tokens
—
No hosted API on OpenRouter; self-host
Context window
1M
tokens

Benchmarks

Each bar shows this model's score. The tick marks the best large model on that benchmark. Sources: 10 reported by Ricursive Intelligence. How we label sources.

MMLU-Pro Lab-reported Knowledge84.2
GPQA Diamond Lab-reported Knowledge85.2
Class best: Nemotron 3.5 Bolt 74B-A9B 88.8
AIME Lab-reported Math · AIME 2025 (AIME 2026: 95.0)98.3
SWE-bench Verified Lab-reported Coding76.5
Terminal-Bench Lab-reported Agentic · Terminal-Bench 2.1 (Terminus 2)80.5
IFBench Lab-reported Instruction following79.2
MMLU Lab-reported Knowledge89.6
HMMT Lab-reported Math · HMMT Feb 202690.5
AA-LCR Lab-reported Long context75.0
SciCode Lab-reported Coding · SciCode (Inspect AI)39.4
Class best: Nemotron 3.5 Bolt 74B-A9B 39.7

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 18
LiveCodeBench≈ 90

Specifications

Lab
Ricursive Intelligence (country not published · Other)
Architecture
Dense transformer
Total parameters
64.8B
Active parameters
All (dense)
Size class
Large
Context window
1M tokens
Input modalities
text
Reasoning
Reasoning (thinking)
License
Apache-2.0
Openness
Open weights
Memory to run
~41.9 GB at 4-bit, 8K context (more sizes)
Hugging Face
15K downloads (30 days) · 615 likes

Compare with

Ranked in

Sources

Something wrong here? Report it.

Run it locally

No GGUF build from a well-known publisher yet.

vLLM (GPU server)
vllm serve ricursiveAI/Ripple-65B-Preview
Full-precision weights, for a server GPU.

Memory needed: weights, context cache and runtime buffers.

Quantization4K context32K context256K context
16-bit F16132 GB134 GB149 GB
8-bit Q8_071 GB73 GB88 GB
4-bit Q4_K_M42 GB43 GB59 GB

Check it against your GPU, Mac or phone →