Primary use stated or inferred cautiously from official documentation.
Hydrogen 1.0 INT8
INT8 distribution of Hydrogen 1.0, a Deep Principle scientific model based on GLM-5.3. The INT8 weights appeared on November 14; this is a new format of the existing model, not a new Hydrogen generation. The verified modality is text.
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.
Limits and integration
| API ID | deep-principle/Hydrogen-1.0-INT8 |
|---|---|
| Model type | Text and reasoning · Code |
| Access model | Downloadable weights |
| License | MIT · component notices retained |
| Deployment | Self-hosted |
| Release | 14 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text |
| Exit | Text |
| Context window | Up to 1M tokens; depends on runtime and resources |
| maximum output | Up to 128K tokens; depends on configuration |
| 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 · SGLang |
Assist with scientific questions, derivations and research planning under specialist review. W8A16 quantizes selected weights, retains activations and other components in BF16 and reduces the checkpoint to about 793 GB. It guarantees neither greater speed nor operation on a home computer.
The team tested SGLang 0.5.20 on eight NVIDIA B300 GPUs. A runtime supporting compressed-tensors pack-quantized is required, along with the supplied config.json and disabled shared-expert fusion. The format is GPU-vendor-neutral, but other hardware is not validated. Review remote code before execution.
What it can do
Orientation
Assist with scientific questions, derivations and research planning under specialist review. W8A16 quantizes selected weights, retains activations and other components in BF16 and reduces the checkpoint to about 793 GB. It guarantees neither greater speed nor operation on a home computer.
Context
Up to 1M tokens; depends on runtime and resources · Up to 128K tokens; depends on configuration
Tools and integration
Not published
Access
Hugging Face · SGLang
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | Not published | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Not published | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
No hosted API price has been verified for this checkpoint. The weights are MIT-licensed and retain component notices; deployment consumes compute and memory. Access to the INT8 repository has been checked without signing in.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
How to read the results
The INT8 card publishes no numerical results specific to this quantization. Scientific demonstrations of the original model combine tools and particular environments; neither their results nor those of GLM-5.3 are transferred to this checkpoint.
i We have not found published benchmarks for this model.
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.
Chronology
deep-principle/Hydrogen-1.0-INT8
New INT8 format of the existing model; access to these weights verified independently of the BF16 repository.