Model library / rockmatrix-ai
EdgeRock-4B is a 4.6 billion-parameter multimodal language model with constant-memory reasoning capabilities.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
EdgeRock-4B is a multimodal model released by rockmatrix-ai. The publisher describes it as providing strong text and image reasoning while keeping a fixed memory budget, using recurrent state and Rock Memory Anchors to summarise distant context. It targets edge devices where memory must be shared between perception, control and inference.
The model has 4.6 billion parameters stored in bfloat16 format and supports a 262 144-token context window. It runs under an "other" licence. In practice it uses about 72 MiB of attention KV per sequence and 97.76 MiB of persistent history, which is 99.1 % smaller than a comparable Qwen3.5-4B model. On a Hopper 80 GB GPU it decodes at 233.7 tokens per second with a 128 K input and output length.
Specification
| Repository | rockmatrix-ai/EdgeRock |
| Publisher | rockmatrix-ai |
| Published | 7 October 2026 |
| Architecture | EdgeRockForCausalLM |
| Model type | edgerock |
| Parameters | 4,578,061,568 |
| Layers | 32 |
| Hidden size | 2,560 |
| Attention heads | 16, 4 key/value heads |
| Context length | 262,144 tokens |
| Vocabulary | 248,320 tokens |
| Weight format | bfloat16 |
| Licence | other |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 8.5 GiB | Full precision as released |
| fp8 | 4.3 GiB | 8-bit, near-lossless on most models |
| int4 | 2.1 GiB | 4-bit, smallest footprint |
The cache costs 128 KiB per token at 16-bit, so the full 262,144-token context of one request needs about 32.0 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 1 | 10.6 GiB | about 86,611 tokens |
| RTX 5090 32GB | 1 | 17.8 GiB | about 145,593 tokens |
| L40S 48GB | 1 | 32.2 GiB | the full 262,144 tokens |
| A100 80GB | 1 | 61.0 GiB | the full 262,144 tokens |
| H100 80GB | 1 | 61.0 GiB | the full 262,144 tokens |
| RTX PRO 6000 96GB | 1 | 75.4 GiB | the full 262,144 tokens |
| H200 141GB | 1 | 115.9 GiB | the full 262,144 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve rockmatrix-ai/EdgeRock \ --max-model-len 86016 \ --gpu-memory-utilization 0.90
Questions
The weights take about 9 GiB at the released precision. That fits on one RTX 4090 24GB, which leaves roughly 11 GiB for the key-value cache.
The repository declares other. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 262,144 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email hello@aiqre.com with the model name and roughly what volume you expect.
Also in the library
Granite Switch 4.1 8B is a modular 8.8 billion-parameter language model that combines a base LLM, decision and safety adapters, a…
EdgeRock-4B is a 4.6 billion-parameter multimodal language model with constant-size memory for text and image reasoning.
Mr.Chatterbox-2.7B-Chat is a 2.7 billion-parameter Llama-style chat model trained on 19th-century British text to simulate a Vict…
ALoDLM-8B is an 8.2 billion-parameter diffusion language model from Amazon for research on adaptive computation, maths, and code …
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.