Primary use stated or inferred cautiously from official documentation.
R
RICURSIVE INTELLIGENCE / ricursiveAI/Ripple-65B-Base-Preview
Ripple-65B Base Preview
A 65-billion-parameter base checkpoint of the Ripple family included in the November 4 release. It is the model after pretraining on 12 trillion tokens, without the later instruction or long-context adaptation stages. It serves as a starting point for research and further training.
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-Base-Preview |
|---|---|
| Model type | 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 |
| Context window | Pretraining at 8K; useful long-context capability unverified |
| maximum output | Not published |
| 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.
Configuration declares up to 2,097,152 positions, but that figure does not demonstrate useful capability at that length in the base checkpoint. Pretraining uses 8K sequences. The post-trained assistant’s 1M context and agent results are not assigned to it. It is not a direct substitute for an instruction-tuned chat model.
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
Pretraining at 8K; useful long-context capability unverified · Not published
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
Input modalities
Text
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
The weights require your own infrastructure. Cost depends on memory, precision, sequence length and load. Review the license and runtime code before deployment. A published performance figure does not guarantee the same cost or speed on another machine.
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
“
The base model’s own table reports 83.7 on five-shot MMLU multiple-choice accuracy. This setup is preserved and the results are not mixed with the assistant’s. Evaluation on conversations, tools and the deployment domain remains necessary.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| MMLU (base, 5-shot MC) | 83,7 / 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 Base Preview
Version added to Inferama's verified catalog.
07 / SOURCES