Up to date
tencent Apache 2.0 out 59 days ago
780B parameters (8 of 256 experts per token), text in, text out, 1M tokens of context. 5 providers serve it from $0.75 per million input tokens.
Verified 2026-10-25 Rank #306 in LLMs · score 83.5 method v1 Sources 4 Downloads 30d 17.6k · #466 of 619 tracked
Picked by rule from 18 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Hy4 preview is a 770B-parameter model with 49B active parameters and a 1M-token context window.
Tencent released Hy4 Preview, a new open weight text-only LLM with 770B total parameters and 49B active parameters.
The 4 picks above, plus these 14. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
5 providers with a live price, from $0.75 to $0.83 per million input tokens; the median is $0.83. One dot per provider, cheapest in teal.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Widely served (5 API providers); 7 quantized builds for local use.
Weights only, an estimate: leave room for the KV cache. Method
About 468 GB at INT4, 936 GB at FP8, 1.87 TB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-08-27), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
9 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | HYV4ForCausalLM |
| Model type | hy_v4 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 779,930,197,312 · F32: 30,795,421 |
| Size band | XXL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 1,048,576 (config.json max_position_embeddings) |
| Papers | arXiv:2512.02556, arXiv:2603.12201 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · HYV4ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for Hy4-preview: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.87 TB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 936 GB · Cluster (≥640 GB) |
| INT4 (0.5 bytes/param) | 468 GB · Multi-GPU node (<640 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.
quantized: 7finetune: 2
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| 6block/Hy4-preview-GGUF | quantized | GGUF | — | — | apache-2.0 | 6block | 1 | 2k |
| inferencerlabs/Hy4-preview-MLX-Q4i | quantized | MLX | — | — | — | inferencerlabs | 0 | 2k |
| qtum/Hy4-preview-GGUF | quantized | GGUF | — | — | apache-2.0 | qtum | 3 | 1.4k |
| mlx-community/Hy4-preview-4bit | quantized | -4bit | 770B | XXL | apache-2.0 | mlx-community | 2 | 1.3k |
| Oxmiq/Hy4-preview-NVFP4-W4A16 | quantized | NVFP4 | 780B | XXL | apache-2.0 | Oxmiq | 2 | 1.1k |
| avar6/Hy4-Preview-gguf | quantized | gguf | — | — | — | avar6 | 0 | 714 |
| anemll/Hy4-preview-FlashMoE-STQ1_0 | quantized | GGUF | — | — | apache-2.0 | anemll | 1 | 425 |
| tiny-random/hy4-preview | finetune | — | 10.9M | S | — | tiny-random | 0 | 19 |
| yujiepan/hy4-preview-tiny-random | finetune | — | 10.9M | S | — | yujiepan | 0 | 19 |
| Date | Commit | SHA |
|---|---|---|
| 2026-08-28 | update | 705d81e |
| 2026-08-28 | update | ce1ba0b |
| 2026-08-28 | update | e985be8 |
| 2026-08-28 | update | ac1efab |
| 2026-08-28 | update | 215d117 |
| 2026-08-28 | Upload folder using huggingface_hub | 3595342 |
| 2026-08-28 | Upload folder using huggingface_hub | 0aa562b |
| 2026-08-28 | Upload folder using huggingface_hub | 79c3855 |
| 2026-08-28 | Update transformers_version to 5.16.2 | 5a05584 |
| 2026-08-28 | Update transformers_version to 5.16.2 | 6e95c4e |
| 2026-08-28 | Update transformers_version to 5.17.0 | c744f4b |
| 2026-08-28 | Update transformers_version to 5.17.0 | 910e6dd |
| 2026-08-28 | Update transformers_version to >=5.16.1 | 2e275d1 |
| 2026-08-28 | Update transformers_version to >=5.16.1 | 5cdbd68 |
| 2026-08-28 | Add .gitignore | 3a88b69 |
| 2026-08-27 | Upload folder using huggingface_hub | 15d5a3d |
| 2026-08-27 | initial commit | 7a996c9 |
Hosted endpoints observed for Hy4-preview, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.75/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Tencent openrouter | tencent/hy4-preview | 1M | fp8 | $0.75 | $2.25 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | tencent/Hy4-preview | 1M | fp4 | $0.79 | $2.38 | 48.7 | 1235 ms | yes | live | 2026-10-25 |
| Novita hf, openrouter | tencent/hy4-preview | 977K | fp8 | $0.83 | $2.50 | 24.1 | 2565 ms | yes | live | 2026-10-25 |
| Cohere openrouter | tencent/hy4-preview | 256K | unknown | $0.83 | $2.50 | — | — | yes | live | 2026-10-24 |
| SiliconFlow openrouter | tencent/hy4-preview | 1M | fp8 | $0.83 | $2.50 | — | — | yes | live | 2026-10-24 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:37 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:06 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
