[whatmodel.sh](../../README.md) / Hy3 preview (295B-A21B) # Hy3 preview (295B-A21B) Tencent · released 2026-04 · 295B params (21B active) · tencent-hy-community **score 66.3** ±1.0 · #120 of 732 · from 24 benchmark results across 2 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Hy3 preview (295B-A21B) among every model on that task or test, worst to best. 
task            vs the others          rank  counts  tests
agent tasks     ▁▁▃▆█▄▅●▄▄▂▁     #91 of 313     38%     11
coding          ▁▁▁▃▅▇█▇●▄▃▁    #116 of 469     33%      4
chat & writing  ▁▁▁▃▆▄▆█▆●▇▃   #108 of 402*     10%      1
math            ●····█·····█        #3 of 3     10%      3
web search      ●·······█··█        #3 of 3    9.1%      4
agent tasks · #91 of 313 · 38% of its score Beats [Grok 4.20 (0309 reasoning)](grok-4-20.md), [GPT-5.1](gpt-5-1.md), [o3-pro](o3-pro.md) here, though they rank higher overall. 
test                                result  vs the others in its run    difficulty            counts  why
SWE-bench Pro (split unspecified)…   46.0%  ▂·▁▁▁▁▂▃●▇▂▁ beats 26/58    very hard · avg 8.3%    6.2%  sharp · at its level · small run
SWE-bench Verified (maker)           74.4%  ▁▁▁▁▁▁▂▄█●▁▁ beats 101/138  hard · avg 24%          5.3%  sharp · at its level · small run
Terminal-Bench 2.1 (maker)           58.0%  ▁·▁·▁▁▄▃●█▅█ beats 25/63    hard · avg 18%          5.1%  sharp · at its level · small run
SWE-bench Multilingual (maker)       68.3%  ▁··▁▃▁▅●█▂▂▁ beats 35/65    hard · avg 29%          3.8%  at its level · small run
APEX-Agents (pass@1) (maker)         12.4%  ▃▃··▃··●▃·▃▃ beats 3/8      very hard · avg 3.1%    3.4%  small run
MCP-Atlas (public set) (maker)       66.1%  ▂·▁▂▂·▁▃●▆▆▂ beats 20/39    hard · avg 33%          3.1%  at its level · small run
Humanity's Last Exam with tools (…   35.4%  ▁·▁▂▂▁●▄▇█▄▅ beats 17/62    very hard · avg 15%     2.6%  at its level · small run
Toolathlon (Tool Decathlon) (make…   32.1%  ▂▂▄▂●▄▆▂█▆██ beats 6/27     very hard · avg 13%     2.4%  at its level · small run
DeepSearchQA (maker)                 82.8%  ▄·▄▄···●▄··█ beats 3/6      medium · avg 57%        2.3%  small run
WideSearch (maker)                   67.8%  █·▂●··▂···▂▂ beats 7/10     medium · avg 40%        2.1%  at its level · small run
BrowseComp with context managemen…   67.1%  ▄·●▂·█▂·▆▄·▂ beats 2/15     medium · avg 44%        2.1%  at its level · small run
coding · #116 of 469 · 33% of its score Beats [Grok 4.3](grok-4-3.md), [Qwen3.5 Plus](qwen3-5-plus.md), [DeepSeek V3.2](deepseek-v3-2.md) here, though they rank higher overall. Trails [Llama 3.1 Nemotron Ultra 253B v1](llama-3-1-nemotron-ultra-253b.md), [Solar Pro 4](solar-pro-4.md), [Nemotron Cascade 2 30B-A3B](nemotron-cascade-2-30b-a3b.md) here, though they rank lower overall. 
test                                result  vs the others in its run    difficulty            counts  why
Arena WebDev (Code Arena)           1356.9  ▁▁▃▂●█▃▆▅▅▁▂ beats 29/129   avg 981                  18%  sharp
SWE-bench Pro (split unspecified)…   46.0%  ▂·▁▁▁▁▂▃●▇▂▁ beats 26/58    very hard · avg 8.3%    6.2%  sharp · at its level · small run
SWE-bench Verified (maker)           74.4%  ▁▁▁▁▁▁▂▄█●▁▁ beats 101/138  hard · avg 24%          5.3%  sharp · at its level · small run
SWE-bench Multilingual (maker)       68.3%  ▁··▁▃▁▅●█▂▂▁ beats 35/65    hard · avg 29%          3.8%  at its level · small run
chat & writing · #108 of 402 · 10% of its score Beats [LongCat 2.0 (1.6T-A48B)](longcat-2-0.md), [Inkling Small (266B-A12B)](inkling-small.md), [GPT-5.4 nano](gpt-5-4-nano.md) here, though they rank higher overall. Trails [Mistral Large 3 (675B-A41B)](mistral-large-3.md), [Gemma 4 12B](gemma-4-12b.md), [GPT-4.1](gpt-4-1.md) here, though they rank lower overall. 
test            result  vs the others in its run    difficulty  counts  why
LMArena (text)  1408.8  ▁▁▁▂▃▃▃▆▆●█▃ beats 253/385  avg 1260       10%  blunt
math · #3 of 3 · 10% of its score 
test                             result  vs the others in its run  difficulty            counts  why
IMO-AnswerBench (maker)           84.3%  ▁▁·▁▁▁▄▆█●▂▃ beats 34/44  medium · avg 44%        5.0%  small run
USAMO 2026 (maker)                37.3%  ●·▄▄▄▄·····█ beats 0/6    very hard · avg 13%     3.4%  sharp · at its level · small run
MathArena Apex (pass@1) (maker)   12.6%  ●····▄···▄█▄ beats 0/5    very hard · avg 8.9%    2.0%  at its level · small run
web search · #3 of 3 · 9.1% of its score 
test                                result  vs the others in its run  difficulty           counts  why
Humanity's Last Exam with tools (…   35.4%  ▁·▁▂▂▁●▄▇█▄▅ beats 17/62  very hard · avg 15%    2.6%  at its level · small run
DeepSearchQA (maker)                 82.8%  ▄·▄▄···●▄··█ beats 3/6    medium · avg 57%       2.3%  small run
WideSearch (maker)                   67.8%  █·▂●··▂···▂▂ beats 7/10   medium · avg 40%       2.1%  at its level · small run
BrowseComp with context managemen…   67.1%  ▄·●▂·█▂·▆▄·▂ beats 2/15   medium · avg 44%       2.1%  at its level · small run
other tests (13) 
test                                result  vs the others in its run    difficulty            counts  why
GPQA Diamond (maker)                 87.2%  ▁▁▁▁▂▄▅▅██●▄ beats 250/322  hard · avg 44%          7.2%  sharp · small run
Humanity's Last Exam (maker)         30.0%  ▁▁▁▄▆▆▇█●▆▁▃ beats 75/111   very hard · avg 6.2%    3.7%  at its level · small run
NL2Repo (maker)                      35.3%  ▂▂▂·▃▃●█▃▂▅▃ beats 7/22     very hard · avg 12%     3.0%  at its level · small run
Claw-Eval Pass^3 (maker)             55.0%  █▄▄·●·▄▄·▄·▄ beats 4/8      hard · avg 29%          2.0%  at its level · small run
SuperChem (maker)                    47.0%  ●··········█ beats 0/1      hard · avg 22%          1.9%  at its level · small run
SkillsBench 79-task text-only (Cl…   29.1%  ●··········█ beats 0/1      hard · avg 16%          1.7%  at its level · small run
FrontierScience-Research (maker)     19.0%  ●█·······█·█ beats 0/3      very hard · avg 7.2%    1.5%  small run
CL-bench (maker)                     22.8%  █···▄······● beats 3/4      very hard · avg 8.4%    1.3%  blunt · small run
WildClawBench (35 text-only queri…   45.3%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …
FrontierScience-Olympiad (maker)     70.0%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …
PHYBench (maker)                     71.4%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …
CMT-Benchmark (maker)                19.4%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …
CL-bench life (maker)                15.7%  ●··········█ beats 0/1                                 —  not counted yet: no other model in the …
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its *counts* column above; here it is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
tencent/Hy3 model card (submitted / mak…       22   71% ▇▇▇▇▇▇▇▇▇
LMArena leaderboard dataset                     2   28% ▇▇▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 66.3
Licences: tencent/Hy3 model card ([Apache-2.0](../../licenses/tencent--Hy3.txt)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/))
Not counted (5) 
source                  test                            score  why
tencent/Hy3 model card  WildClawBench (35 text-only q…  45.3%  too few models have it yet
tencent/Hy3 model card  FrontierScience-Olympiad        70.0%  too few models have it yet
tencent/Hy3 model card  PHYBench                        71.4%  too few models have it yet
tencent/Hy3 model card  CMT-Benchmark                   19.4%  too few models have it yet
tencent/Hy3 model card  CL-bench life                   15.7%  too few models have it yet
## How to run it ### Other ways to run it 1. Run it as published (BF16, needs ~597.58 GB, NVIDIA only, from [huggingface.co/tencent](https://huggingface.co/tencent/Hy3-preview/blob/549c2b3a0fd5b9a6c6059a9935bf0d59ab69d75a/README.md)) ```sh vllm serve tencent/Hy3-preview --tensor-parallel-size 8 --speculative-config.method mtp --speculative-config.num_speculative_tokens 1 --tool-call-parser hy_v3 --reasoning-parser hy_v3 --enable-auto-tool-choice --served-model-name hy3-preview ``` Model repository and documentation: [huggingface.co/tencent](https://huggingface.co/tencent/Hy3-preview) [interactive version](https://whatmodel.sh/model/hy3-preview) · [JSON](https://whatmodel.sh/api?format=json&all=1)