Continued light-mode training on glasseye RTX 5070 only (no Modal/HF Jobs). Load with MoE?Expert swap as documented on the baseline card.
Runs On
What it takes to serve opus-4.8-recreation-1b-light-v4 (929M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 1.9 GB | 2.2 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.9 GB | 1.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.5 GB | 0.6 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.
Model Card
By 0ai, published under mit, revision 07030291618f.
Continued light-mode training on glasseye RTX 5070 only (no Modal/HF Jobs). Load with MoE?Expert swap as documented on the baseline card.
Read 0ai's full model card
Opus 4.8 Recreation 1B Light ? v4 local refresh
Continued light-mode training on glasseye RTX 5070 only (no Modal/HF Jobs).
- Init:
cyberviser/opus-4.8-recreation-1b-light/GLASSEYE/opus-4.8-recreation-1b-light - Steps: 150 | light=True | n_loops=1 | final_loss ? 2.12
- Hardware: NVIDIA GeForce RTX 5070 (~12 GB)
Load with MoE?Expert swap as documented on the baseline card.
Developed by: cyberviser / GLASSEYE (CyberviserAI)
Configuration
- Vocabulary size
- 199,998
- RoPE base
- 500000
Identity and Version
- Repository
- cyberviser/opus-4.8-recreation-1b-light-v4
- Publisher
- 0ai
- Task
- Not stated by the source
- Modality
- Other
- Library
- open_mythos
- Parameters
- 929M parameters
- Languages
- Not stated by the source
- Revision
- 07030291618f29b09e4b044a8fa5db6ab6e3c133
- First published
- 2026-09-18
- Last updated
- 2026-09-18
Files and Weights
6 files, 7.4 GB in total. The weights are 2 files totalling 7.4 GB in bin, safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 3.7 GB | 66bd0557f686 |
| pytorch_model.bin | Weights | 3.7 GB | 83703143c5bc |
| build_metadata.json | Configuration | 265 B | — |
| config.json | Configuration | 514 B | — |
| README.md | Documentation | 591 B | — |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- mit
- Access
- Open weights, no gate
- Download size
- 7.4 GB
Released by 0ai through its official repository on Hugging Face. Read the license.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 7.4 GB |
| 16-bit | 1.9 GB |
| 8-bit | 0.9 GB |
| 4-bit | 0.5 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About opus-4.8-recreation-1b-light-v4
How much GPU memory does opus-4.8-recreation-1b-light-v4 need?
About 2.2 GB at 16-bit and 0.6 GB at 4-bit: the weights (929M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run opus-4.8-recreation-1b-light-v4 on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use opus-4.8-recreation-1b-light-v4 commercially?
Yes. opus-4.8-recreation-1b-light-v4 is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.