Primary use stated or inferred cautiously from official documentation.
Tencent Hy4 Base
Hy4 pretrained checkpoint with weights published on November 20. Its language backbone has 770 billion parameters and activates 49 billion per token; vision supports images and video. It is distinct from the Hy4 assistant and its FP8 quantization.
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 | tencent/Hy4-base |
|---|---|
| Model type | Multimodal · Text and reasoning · Code |
| Access model | Open weights |
| License | Tencent Hy License 1.0 |
| Deployment | Self-hosted |
| Release | 20 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image · Video |
| Exit | Text |
| Context window | 1.048.576 tokens |
| maximum output | Text; separate output maximum unpublished |
| 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 · Transformers |
Research and further adaptation; not an instruction-tuned chat model.
Tencent Hy License has its own commercial conditions for large MaaS providers; it is not Apache 2.0. Weight context does not replace TokenHub limits. Synthetic data and answer formats can affect the few-shot table.
What it can do
Orientation
Research and further adaptation; not an instruction-tuned chat model.
Context
1.048.576 tokens · Text; separate output maximum unpublished
Tools and integration
Not published
Access
Hugging Face · downloadable checkpoint
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 |
No verified official API tariff. Self-hosting requires memory, compute and maintenance; downloadable weights do not make inference free.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
How to read the results
Hy4 Base scores 73.9% on MMLU-Pro with five examples and macro accuracy. This Tencent evaluation concerns the base model, not the assistant or FP8.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| MMLU-Pro (Hy4 Base 5-shot) | 73.9 % | Macro accuracy · 5-shot | 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.
Chronology
tencent/Hy4-base
BF16 checkpoint.