Hunyuan Hy3
Tencent's large MoE, ranked seventh among open-weight models in September 2026. The official weights contain about 299B parameters and the model card reports 21B activated per token. Note the community licence carries geographic restrictions that matter for EU commercial use.
TencentMixture of expertsgeneralreasoning7 machines can run it
Total parameters
298.786B
Official spec sets memory need
Active parameters
21B
Official spec sets decode speed
Context
256K
Official spec tokens
4-bit weights
163 GB
Calculated before KV cache
This is a sparse mixture of experts. All 298.786B of weights must be available to the runtime, but only 21B are read per generated token. Ordinary runtimes keep the full quantized model in memory; a runtime-specific SSD-streaming artifact can retain a smaller working set and fetch expert data on demand, trading speed for capacity.
Model specification
Publisher
Tencent Official spec
Architecture
Sparse mixture of experts Official spec
Total parameters
298.786B Official spec
Active parameters
21B per token Official spec
Context length
262,144 tokens Official spec
Attention ⓘ
GQA · 8 KV heads × 128 dim × 80 layers Official spec
Licence
Tencent Hunyuan Community Official spec source
Specification confidence ⓘ
High · verified 2026-10-19 Official spec source
Released
1 Jun 2026 Official spec
Official source
Quantizations and memory
| Quantization | Format | Bits/weight | Resident | Download | Quality kept |
|---|---|---|---|---|---|
MLX 4-bit | mlx | 4.5 | 162.8 GB | 162.8 GB | 98.0% |
Q4_K_Mdefault | gguf | 4.85 | 173.8 GB | 173.8 GB | 98.5% |
Q5_K_M | gguf | 5.69 | 203.9 GB | 203.9 GB | 99.3% |
Q8_0 | gguf | 8.5 | 301.6 GB | 301.6 GB | 99.9% |
Generic weight sizes are computed from the parameter count and bits per weight plus a format-specific overhead. Runtime-specific artifacts use their published resident and download footprints; streamed models can therefore require much more disk than memory. Quality retention is an assumption, not a measured evaluation.
Coding & quality benchmarksCompare coding results →
No published benchmark results have been imported for this model yet. This is missing evidence, not a score of zero.
We show source metrics rather than deriving one opaque quality number. Different benchmarks measure genuinely different things, and collapsing them into a single score would hide exactly the disagreements worth seeing.
Recommendations
Cheapest that can run it
~32.9–47.3 t/s · $5,674
Lowest purchase price among configurations where the model fits at some quantization in our catalogue. Speed is not considered.
Cheapest above 20 t/s
~32.9–47.3 t/s · $5,674
20 tokens/second is roughly the point at which generation keeps pace with reading. Below it, interactive use feels like waiting.
Cheapest above 40 t/s
~32.9–47.3 t/s · $5,674
40 tokens/second is the threshold most people describe as comfortable for coding agents, where output arrives faster than you can review it.
Fastest with real measurements
Nothing in the database qualifies.
Highest throughput among configurations with an actual published measurement rather than our estimate.
Best throughput per purchase-price unit
~57.7–83 t/s · $8,647
Highest decode tokens/second per 1,000 units of the displayed purchase currency. Ignores running costs and resale — see the economics section for the full picture.
Hardware that runs Hunyuan Hy30 measured, 7 estimated
7 of 7 rows
| Hardware↕ | Quant | Memory↕ | Decode▼ | Prefill↕ | Context | Price↕ | Fit | Confidence↕ |
|---|---|---|---|---|---|---|---|---|
| NVIDIA DGX H200 (8x H200, 1,128 GB) NVIDIA · 1,128 GB · 38,400 GB/s | Q8_0recommended | 316.9 GB | ~347–500 t/s | ~33780–70150 t/s | 256K | $359,430 | Comfortable | Estimated |
| Lenovo ThinkStation PX (4x RTX PRO 6000, 384 GB) Lenovo · 384 GB · 7,168 GB/s | Q5_K_Mrecommended | 214.7 GB | ~103–148 t/s | ~1620–3370 t/s | 256K | $58,553 | Comfortable | Estimated |
| Mac Studio M5 Ultra 256 GB Apple · 256 GB · 1,200 GB/s | MLX 4-bitrecommended | 171.2 GB | ~57.7–83 t/s | ~366–761 t/s | 128K | $8,647 | Comfortable | Estimated |
| Mac Studio M3 Ultra 256 GB Apple · 256 GB · 819 GB/s | MLX 4-bitrecommended | 171.2 GB | ~32.9–47.3 t/s | ~88.4–184 t/s | 128K | $5,674 | Comfortable | Estimated |
| NVIDIA DGX Station GB300 (748 GB) NVIDIA · 748 GB · 7,100 GB/s | Q8_0recommended | 314.1 GB | ~32–46 t/s | ~4510–9370 t/s | 256K | $90,082 | Comfortable | Estimated |
| Mac Studio M5 Ultra 512 GB Apple · 512 GB · 1,200 GB/s | Q8_0recommended | 314.1 GB | ~31.6–45.4 t/s | ~366–761 t/s | 256K | $11,350 | Comfortable | Estimated |
| Mac Studio M3 Ultra 512 GB Apple · 512 GB · 819 GB/s | Q8_0recommended | 314.1 GB | ~17.7–25.5 t/s | ~88.4–184 t/s | 256K | $8,287 | Comfortable | Estimated |
Submit a benchmarkContributions are reviewed before publication.
Similar models
MiMo-V2.5 310B-A15B
XiaomiMiMo · 310.775B (15B active) · MIT
DeepSeek V2 236B-A21B
DeepSeek · 235.741B (21B active) · DeepSeek
DeepSeek V2.5 236B-A21B
DeepSeek · 235.741B (21B active) · DeepSeek
Qwen3 235B-A22B
Alibaba Qwen · 235.094B (22B active) · Apache 2.0
MiniMax Text 01 456B-A46B
MiniMax · 456.09B (46B active) · MiniMax
Mixtral 8x22B
Mistral AI · 140.63B (39B active) · Apache 2.0