- Type
- Open weightsApache License 2.0
- Params
- 299B
- Context
- 262K
active per word not recorded by us · about 197K words of context
Our take
Written Sep 17, 2026Hy3 is a large downloadable text model from Tencent, released under a licence that allows commercial use, changes and redistribution. Its published scores come from human-preference arenas and agent sessions rather than task-accuracy tests, so they say which answers people preferred, not whether the work was done right.
Use it for everyday text work where you want a large model you can download and build on, or for long-document work where the request capacity means material need not be split up first. Treat the published scores as preference and relative ranking rather than correctness, and trial it on work you can check yourself. Skip it if your task needs a measured accuracy figure, or if you meant to run a model of this size on a single workstation.
The case for it
- The licence allows commercial use, changes and redistribution (Apache License 2.0), and the weights are downloadable, so a product can be built on it without a licence negotiation.
- 262144 tokens of request capacity means long documents need not be split up first, though reliable recall across all of it is unverified in our data.
- Its coding and web-app building arena scores sit above its other supplied arena figures, but those record which answer people preferred, not whether the code was correct.
The case against it
- No accuracy benchmark is supplied: every score is a preference or relative ranking, so a trial on work you can check yourself is the only way to judge it.
- Agent-session results are its weakest measured area, with task outcome and overall among its lowest supplied agent scores on a scale where higher is better.
- 298.8 billion parameters with no per-token active figure supplied, so the whole model is in play on every token and the hardware requirement is not reduced by any sparsity.
How good is it?
An open text model for everyday questions, though calling tools to carry out requests is where it struggles.
- getting answers to everyday questionsArena Text (overall) · 41st of 168
- calling tools to carry out requestsArena Agent · Tool use · 53rd of 55
EverydayGeneral questions and everyday reasoning
Arena Text (overall)41st of 168 · 1457
CodingWriting and fixing code on its own
Arena Coding50th of 168 · 1501
AgenticPlanning, calling tools, staying on task
Arena Agent41st of 55 · −0.056
Arena Agent is the only board that has scored it for this.
WritingDrafting and rewriting prose
Arena Creative Writing51st of 168 · 1419
Arena Creative Writing is the only board that has scored it for this.
Placings on Arena's agent boards, from live sessions people ran themselves. A model can lead on one of these and sit mid-field on the others.
Boards this model appears on that none of the ratings above are built on.
Every published score for this model12 scoresEvery figure we hold, from 12 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M3 Ultra (80-core GPU) · 512 GB
Room to spare. 188.1 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 4 hours and 10 hours ago — each listing carries its own date.
- per 1M tokens
- $0.13 in / $0.53 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.13 / $0.53checked 10 hours ago | 262K | not measured | Unknown | Unknown | Unknown |
| Tencentfp8Through OpenRouter | $0.13 / $0.53checked 10 hours ago | 262K128K max reply | 57 tok/s | No | No | Confirmed |
| DeepInfrafp4Direct and through OpenRouter | $0.13 / $0.53checked 4 hours ago | 262K131K max reply through OpenRouter | 99 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Novita AIDirect and through OpenRouter | $0.14 / $0.58checked 4 hours ago | 262K236K max reply through OpenRouter | 60 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| GMICloudbf16Through OpenRouter | $0.14 / $0.58checked 4 hours ago | 262K236K max reply | 24 tok/s | No | Yesunknown period | Unknown |
| PhalaThrough OpenRouter | $0.15 / $0.64checked 4 hours ago | 262K236K max reply | 34 tok/s | No | No | Confirmed |
| AtlasCloudfp8Through OpenRouter | $0.20 / $0.80checked 4 hours ago | 262K131K max reply | 27 tok/s | No | Yesunknown period | Unknown |
DeepInfra
Novita AIAcross the 7 listings we hold: 6 say they do not train on prompts (2 of them only through OpenRouter), 0 say they do and 1 does not say. 4 appear in the zero-retention registry we check (2 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| Tencentfp8Through OpenRouter | ✓ | ✓ | ✗ |
| DeepInfrafp4Direct and through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIDirect and through OpenRouter | ✓ | ✗ | ✓ |
| GMICloudbf16Through OpenRouter | ✓ | ✓ | ✗ |
| PhalaThrough OpenRouter | ✓ | ✗ | ✓ |
| AtlasCloudfp8Through OpenRouter | ✓ | ✗ | ✓ |
Tool calling: 7 of 7 listings say yes. JSON output: 4 of 7 listings say yes, 3 say no. Strict schema: 5 of 7 listings say yes, 2 say no.
Models people weigh against Hy3
When we formed this view
Recent changes
Each date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- We do not hold the active parameter count for it, so how much of it runs on any one token is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 7 listings does not say whether it trains on prompts, and 2 answer only through OpenRouter, not for their own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsApache License 2.0, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Apache License 2.0
Fully permissive: commercial use, redistribution, and derivatives allowed. Requires attribution and a copy of the license. Includes an express patent grant.
Identifiers
- Hugging Face
- tencent/Hy3
- Architecture
- Mixture of experts
- Takes in, gives back
- Text in, text out
- Catalogue slug
- tencent-hy3