whatmodel.sh / Hy4

Hy4 is a closed-weights language model from Tencent, released November 2026. Licence: proprietary.

Catalogue identity notes

Separate released API model, not hy4-preview: TokenHub added 'Hy4' (model parameter hy4) on 2026-11-04 (product updates: '新增支持模型 | 新增支持 Hy4 模型。 | 2026-11-04'), with image understanding that the hy4-preview row lacks; Tencent's 2026-11-09 notice retires hy4-preview and tells users to migrate to 'Hy4 模型(model 参数值:hy4)', auto-switching remaining users from 2026-12-07. Weights unpublished; parameters for the stable model not stated by Tencent. Corresponds to Arena agent-board entry 'Hy4' (org 'tencent', Proprietary); the separate Arena label 'Hy4 preview' stays on hy4-preview. Licence: API-only as of 2026-11-12: Tencent's Hugging Face org has tencent/Hy4-preview and tencent/Hy4-preview-FP8 but no Hy4 repo (HF API search 'Hy4', author=tencent). Third-party news says Tencent promised Hy4 weights 'in the coming days'; not found on a Tencent page, so not recorded. Arena agent board lists 'Hy4' as Proprietary. List price not recorded (USD per 1M tokens from models.dev tencent-tokenhub/hy4.toml (cache read $0.042 excluded). REVIEW: the toml comment says it carries the Hy4 preview listing converted from CNY because no separate Hy4 row existed when written. Tencent's own China-site price page (https://cloud.tencent.com/document/product/1823/130055, observed 2026-11-12) now has a Hy4 row 'Hy4 | hy4 | - | - | 6 | 18 | 0.3' CNY per 1M tokens (input/output/cache hit), identical to Hy4 preview, so the converted USD figures match the official CNY list price. A limited-time 50% discount on Hy4 until 2026-12-31 23:59:59 Beijing time is excluded. The international USD page (tencentcloud.com/document/product/1300/78937) returned a bot challenge to automated fetching and was not used. Drop the numeric price if the reviewer requires a USD figure printed by Tencent.)

Model information sources

Release date ↗

77.6±1.4#22 of 743 LLMsfrom 3 benchmark results across 2 sources

Score within the LMArena (text) comparison set only.

$ how it compares

The LLMs ranked either side of Hy4, out of 897 with a score, then more from Tencent.

#modelscorewhereRAMreleased
20Muse Wonder Meta78±1api $8.00/M2026-10
21GPT-5.5 OpenAI78±0api $11.25/M2026-04
22Hy4 Tencent78±1api price unknown2026-11
23MiMo-V2.6-Pro Xiaomi77±1api $0.54/M +local~443GB2026-09
24DeepSeek V4.1 Flash 1017 DeepSeek76±1api $0.26/M2026-10
more from Tencent
26Hy4 preview (770B-A49B) Tencent76±1api $1.25/M +local~248GB2026-08
72Hunyuan Hy3 (295B-A21B) Tencent72±1api $0.23/M +local~105GB2026-07
149Hy3 preview (295B-A21B) Tencent66±1local~629GB2026-04

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▄▅▅▃▃▃▁ #19 of 319 · 99% of its score · 3 tests

Beats GPT-5.5 Pro, GPT-5.6 Sol, Claude Mythos Preview (limited preview) here, though they rank higher overall. Trails GLM-5.3, DeepSeek V4.1 Flash 1017 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena Agent (October 2026 aggregate IPS)0.0526 IPS ✎▁▁▁▃▃█▃▁▁▁·▁ beats 49/63avg -0.403 IPS▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 60%sharp
Terminal-Bench 4.039.1% ✎▁····▁▄▇▅▇██ beats 16/30very hard · avg 0.7%▇▇▇▇▇▇▇▇▇▇ 25%sharp · at its level
Terminal-Bench 4.137.9% ✎▂▂▂··▅·▂▅█▃▆ beats 12/20very hard · avg 1.0%▇▇▇▇▇▇ 15%sharp · at its level · noisy

$ where to run it

Through an API

api · Tencent Cloud TokenHub · price unknown · 1M context · cloud.tencent.com ↗

$ how the score was worked out · LMArena leaderboard dataset 60%, Terminal-Bench 4.0 official l… 39%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The 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's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
LMArena leaderboard dataset1CC BY 4.0
Terminal-Bench 4.0 official leaderboard submissions2Apache-2.0
prior · pull toward the middle
weighted average= 77.6

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json