Felo API PlatformFelo API Platform

Ricursiveai API: models, pricing, and specs

Compare every Ricursiveai model available through Felo API — prices, context windows, and capabilities in one table. Call any of them with the same API key.

Models
1
chat ×1
Input per 1M
$0.2
Output per 1M
$0.6
Largest context
1.049M

Ripple-65B Preview is Ricursive Intelligence's open-weight 65-billion parameter dense Transformer for agentic coding, tool use, and technical question answering.

Ripple-65B Preview is the first model from Ricursive Intelligence, released on November 5, 2026 as open weights under the Apache 2.0 license. It is a 65-billion parameter dense Transformer pretrained from scratch on approximately 12 trillion tokens of mostly English text and code. The release includes three checkpoints: an instruction-tuned BF16 model, an NVFP4 version that quantizes MLP layers while keeping attention in BF16 with weights around 34 GiB at 4-bit, and a BF16 base model. The dense architecture was chosen over Mixture-of-Experts to reduce infrastructure complexity and enables running on consumer hardware such as a MacBook or NVIDIA DGX Spark; the BF16 model reaches 250 tokens per second on 2x B200 GPUs at batch size 8 before quantization.

The model accepts text input and produces text output, with a 1 million token context window. It supports native tool calling with one function call per turn, returning function names with JSON arguments, and a reasoning mode that returns internal reasoning as a separate section before the final answer. Optional system instructions, user and session identifiers, function definitions, sampling controls, stop strings, a numeric seed, and a toggle between deterministic decoding and sampling are all available. It handles code review and refactoring plans for files and modules up to about 15,000 tokens.

Ripple-65B Preview fits coding agents that plan changes, call tools, and continue from results, as well as code review and refactoring assistance, technical question answering, assistants that look up data through user-defined functions, and document summarization and cross-checking. Its preview release has undergone relatively light posttraining focused on agentic coding, it supports only one function call per turn, and it trails larger Mixture-of-Experts rivals on SciCode, DeepSWE, and MCP Atlas.

Specifications

Input per 1M
$0.2
Output per 1M
$0.6
Context
1.049M
Max output
131.072K
Protocols
messages
View model →

Similar models from other providers