Primary use stated or inferred cautiously from official documentation.
R
RICURSIVE INTELLIGENCE / ricursiveAI/Ripple-65B-Preview
Ripple-65B Preview
Ricursive Intelligence text and coding assistant released on November 4, 2026. It is a dense Transformer with 65 billion parameters and 80 layers, pretrained on 12 trillion tokens. This profile groups BF16 weights and their NVFP4 variant, with validation specific to each precision.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | ricursiveAI/Ripple-65B-Preview |
|---|---|
| Model type | Code · Text and reasoning |
| Access model | Open weights |
| License | Apache 2.0 |
| Deployment | Self-hosted |
| Release | 04 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text |
| Exit | Text · Code |
| Context window | Up to 1M tokens according to the model card |
| maximum output | 128K tokens on OpenRouter |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · PyTorch |
Research and evaluation with your own data, preserving the exact weight revision, configuration and dependencies. Compare accuracy, memory and latency before integrating the model into an application.
The base model and DFlash2 have separate profiles. The advertised context does not guarantee retention of every detail. Configuration includes custom code, so review the runtime and pin versions. Do not assign BF16 scores to NVFP4 without specific evaluation. A server may impose lower limits than the weights.
03 / CAPABILITIES
What it can do
Orientation
Research and evaluation with your own data, preserving the exact weight revision, configuration and dependencies. Compare accuracy, memory and latency before integrating the model into an application.
Context
Up to 1M tokens according to the model card · 128K tokens on OpenRouter
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
Text
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | 0.20 USD / 1M tokens on OpenRouter | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | 0.60 USD / 1M tokens on OpenRouter | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
OpenRouter lists 0.20 USD input and 0.60 USD output per million tokens for its channel. These are hosting rates, independent of deploying the Apache 2.0 weights. Local cost depends on hardware and load. No Batch discount has been verified for this channel.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
Ricursive reports 80.5 on Terminal-Bench 2.1 with Terminus 2 and 79.1 on SWE-bench Pro v2.0 Hard for the assistant. These are provider figures not reproduced by Inferama. The table also includes base-model results, which must be read separately. They do not establish the standalone performance of the DFlash2 drafter.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| Terminal-Bench 2.1 | 80,5 / 100 | Score | View source ↗ |
| SWE-bench Pro v2.0 Hard | 79,1 / 100 | Score | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Ripple-65B Preview
Version added to Inferama's verified catalog.
07 / SOURCES