Semaphore at Helsby. 24 May 2020. #railway #photography
November 21, 2025 at 7:30 AM
Semaphore at Helsby. 24 May 2020. #railway #photography
New (short!) blog post up: on OpenRouter's AI Model Rankings, I noticed a peculiar new LLM topping the rankings by a large margin: Hy3. I looked into the data and only became more confused. minimaxir.com/2026/05/open...
The mysterious Hy3 LLM is topping OpenRouter Model Rankings by a large margin
Why Hy3?
minimaxir.com
May 26, 2026 at 3:40 PM
New (short!) blog post up: on OpenRouter's AI Model Rankings, I noticed a peculiar new LLM topping the rankings by a large margin: Hy3. I looked into the data and only became more confused. minimaxir.com/2026/05/open...
Tencent's Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.
huggingface.co/tencent/Hy3
huggingface.co/tencent/Hy3
July 6, 2026 at 7:48 AM
Tencent's Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.
huggingface.co/tencent/Hy3
huggingface.co/tencent/Hy3
July 14, 2026 at 8:56 PM
Introducing Hy3
hy.tencent.com/research/hy3...
#Cognitive #Science #Machine
#Learning #DeepLearning #AI
#Research #BigData #Statistics
#Computer #Science #Data
#Analytics #Quantum #Computing
#ComputerVision #NaturalLanguageProcessing
#Robotic #Engineering
hy.tencent.com/research/hy3...
#Cognitive #Science #Machine
#Learning #DeepLearning #AI
#Research #BigData #Statistics
#Computer #Science #Data
#Analytics #Quantum #Computing
#ComputerVision #NaturalLanguageProcessing
#Robotic #Engineering
July 8, 2026 at 4:28 AM
What the heck is this Hy3 model and why does it have so many tokens generated with it? The input token price is half that of DeepSeek V4 Flash, but that doesn't explain why it blew up overnight especially since it's a month old. openrouter.ai/tencent/hy3-...
May 16, 2026 at 10:42 PM
What the heck is this Hy3 model and why does it have so many tokens generated with it? The input token price is half that of DeepSeek V4 Flash, but that doesn't explain why it blew up overnight especially since it's a month old. openrouter.ai/tencent/hy3-...
Hy3 preview (295B A21B and open-weight)
A leading reasoning and agent model in its size, with great cost efficiency.
Model: huggingface.co/tencent/Hy3-...
Blog (In Chinese): hy.tencent.com/hy3-preview
A leading reasoning and agent model in its size, with great cost efficiency.
Model: huggingface.co/tencent/Hy3-...
Blog (In Chinese): hy.tencent.com/hy3-preview
April 23, 2026 at 8:20 PM
Hy3 preview (295B A21B and open-weight)
A leading reasoning and agent model in its size, with great cost efficiency.
Model: huggingface.co/tencent/Hy3-...
Blog (In Chinese): hy.tencent.com/hy3-preview
A leading reasoning and agent model in its size, with great cost efficiency.
Model: huggingface.co/tencent/Hy3-...
Blog (In Chinese): hy.tencent.com/hy3-preview
I still don't understand Hy3 preview and why it's so high though.
June 14, 2026 at 3:52 PM
I still don't understand Hy3 preview and why it's so high though.
Tencent just released HY3 🔥
huggingface.co/collections/...
- 295B / 21B MoE , 256K context
- Apache 2.0
- FP8 version included 👀
- Switchable reasoning: no think / low / high
- Hallucinates half as often as before
- Stable across agent frameworks
huggingface.co/collections/...
- 295B / 21B MoE , 256K context
- Apache 2.0
- FP8 version included 👀
- Switchable reasoning: no think / low / high
- Hallucinates half as often as before
- Stable across agent frameworks
July 6, 2026 at 8:20 AM
Tencent just released HY3 🔥
huggingface.co/collections/...
- 295B / 21B MoE , 256K context
- Apache 2.0
- FP8 version included 👀
- Switchable reasoning: no think / low / high
- Hallucinates half as often as before
- Stable across agent frameworks
huggingface.co/collections/...
- 295B / 21B MoE , 256K context
- Apache 2.0
- FP8 version included 👀
- Switchable reasoning: no think / low / high
- Hallucinates half as often as before
- Stable across agent frameworks
🎵🎶 ✮ ｡ ✩ ｡ ° • ✧ 😊 ✧ • ° ｡ ✩ ｡ ✮ 🎶🎵
#WochenendOhrwuermer
🎹 Piano 🎹
Queen - Bohemian Rhapsody
#EatTheFriedRich #MeinSongDesTages #Cannabis4MyEars
youtu.be/hY3FuBZM7kk
music.youtube.com/search?q=hY3...
#WochenendOhrwuermer
🎹 Piano 🎹
Queen - Bohemian Rhapsody
#EatTheFriedRich #MeinSongDesTages #Cannabis4MyEars
youtu.be/hY3FuBZM7kk
music.youtube.com/search?q=hY3...
Bohemian Rhapsody (Remastered 2011)
YouTube video by Queen - Topic
youtu.be
March 28, 2026 at 4:27 PM
🎵🎶 ✮ ｡ ✩ ｡ ° • ✧ 😊 ✧ • ° ｡ ✩ ｡ ✮ 🎶🎵
#WochenendOhrwuermer
🎹 Piano 🎹
Queen - Bohemian Rhapsody
#EatTheFriedRich #MeinSongDesTages #Cannabis4MyEars
youtu.be/hY3FuBZM7kk
music.youtube.com/search?q=hY3...
#WochenendOhrwuermer
🎹 Piano 🎹
Queen - Bohemian Rhapsody
#EatTheFriedRich #MeinSongDesTages #Cannabis4MyEars
youtu.be/hY3FuBZM7kk
music.youtube.com/search?q=hY3...
"Construire une telle chose est inacceptable"
👉 https://l.sciencesetavenir.fr/HY3
👉 https://l.sciencesetavenir.fr/HY3
October 22, 2025 at 11:30 AM
"Construire une telle chose est inacceptable"
👉 https://l.sciencesetavenir.fr/HY3
👉 https://l.sciencesetavenir.fr/HY3
テンセントが高性能推論モデル「Hy3 preview」を公開、295B-A21BなMoEモデルで高い効率性
https://gigazine.net/news/20260424-tencent-hy3/
https://gigazine.net/news/20260424-tencent-hy3/
テンセントが高性能推論モデル「Hy3 preview」を公開、295B-A21BなMoEモデルで高い効率性
中国の巨大IT企業であるテンセントが開発する大規模言語モデル(LLM)ファミリーの「Tencent HY」から、高性能推論モデル「Hy3 preview」がオープンソースで公開されました。
gigazine.net
April 24, 2026 at 2:32 AM
テンセントが高性能推論モデル「Hy3 preview」を公開、295B-A21BなMoEモデルで高い効率性
https://gigazine.net/news/20260424-tencent-hy3/
https://gigazine.net/news/20260424-tencent-hy3/
Another release today that's probably going to slip under the radar: dots3 (from China's Rednote). MoE (280B, 16B active) that claims parity (roughly) with Tencent's Hy3.
studio.dots.ai/dots/dots3-e...
studio.dots.ai/dots/dots3-e...
小红书
小红书
studio.dots.ai
August 14, 2026 at 8:08 AM
Another release today that's probably going to slip under the radar: dots3 (from China's Rednote). MoE (280B, 16B active) that claims parity (roughly) with Tencent's Hy3.
studio.dots.ai/dots/dots3-e...
studio.dots.ai/dots/dots3-e...
The Hy3 LLM is making waves, leading OpenRouter Model Rankings by a significant margin! Its impressive performance has sparked curiosity and discussion across the AI community. What do you think sets Hy3 apart from the competition? #AI #MachineLearning…
https://minimaxir.com/2026/05/openrouter-hy3/
https://minimaxir.com/2026/05/openrouter-hy3/
July 10, 2026 at 8:06 PM
The Hy3 LLM is making waves, leading OpenRouter Model Rankings by a significant margin! Its impressive performance has sparked curiosity and discussion across the AI community. What do you think sets Hy3 apart from the competition? #AI #MachineLearning…
https://minimaxir.com/2026/05/openrouter-hy3/
https://minimaxir.com/2026/05/openrouter-hy3/
Tencent's AngelSpec, an end-to-end speculative decoding framework supporting both training and deployment.
On Hy3-A21B, DFly delivers a 1.98–2.40× end-to-end speedup over autoregressive decoding across tested concurrency levels from 4 to 64, with 10.5–11.8% higher throughput than DFlash.
On Hy3-A21B, DFly delivers a 1.98–2.40× end-to-end speedup over autoregressive decoding across tested concurrency levels from 4 to 64, with 10.5–11.8% higher throughput than DFlash.
August 3, 2026 at 10:14 PM
Tencent's AngelSpec, an end-to-end speculative decoding framework supporting both training and deployment.
On Hy3-A21B, DFly delivers a 1.98–2.40× end-to-end speedup over autoregressive decoding across tested concurrency levels from 4 to 64, with 10.5–11.8% higher throughput than DFlash.
On Hy3-A21B, DFly delivers a 1.98–2.40× end-to-end speedup over autoregressive decoding across tested concurrency levels from 4 to 64, with 10.5–11.8% higher throughput than DFlash.
Tencent เปิดตัวโมเดล Hy3 ใช้ RL ที่สร้างเอง เปิด API ราคาเริ่มต้น 1.2/4 หยวน
#ShoperGamer #Tencent #Hy3 #Ai #RL #API #ModelAi #Feed
#ShoperGamer #Tencent #Hy3 #Ai #RL #API #ModelAi #Feed
[Shoper Gamer] Tencent เปิดตัวโมเดล Hy3 ใช้ RL ที่สร้างเอง เปิด API ราคาเริ่มต้น 1.2/4 หยวน
โดย
โดย
www.blockdit.com
May 1, 2026 at 2:34 PM
Tencent just released HY3 preview on @hf.co
First drop from TencentHunyuan rebuilt infra👀
Model: huggingface.co/collections/...
Demo: huggingface.co/spaces/tence...
✨ 295B MoE /21B active
✨ 256k context
✨ Hybrid fast/slow thinking
✨ Solid BrowseComp & WideSearch (search agents)
First drop from TencentHunyuan rebuilt infra👀
Model: huggingface.co/collections/...
Demo: huggingface.co/spaces/tence...
✨ 295B MoE /21B active
✨ 256k context
✨ Hybrid fast/slow thinking
✨ Solid BrowseComp & WideSearch (search agents)
April 23, 2026 at 11:05 AM
Tencent just released HY3 preview on @hf.co
First drop from TencentHunyuan rebuilt infra👀
Model: huggingface.co/collections/...
Demo: huggingface.co/spaces/tence...
✨ 295B MoE /21B active
✨ 256k context
✨ Hybrid fast/slow thinking
✨ Solid BrowseComp & WideSearch (search agents)
First drop from TencentHunyuan rebuilt infra👀
Model: huggingface.co/collections/...
Demo: huggingface.co/spaces/tence...
✨ 295B MoE /21B active
✨ 256k context
✨ Hybrid fast/slow thinking
✨ Solid BrowseComp & WideSearch (search agents)
Hello, this is Y-API, a large language model API that everyone can afford.
You can always use DeepSeek v4 flash, mimo v2.5, hy3, and more models in the future for free.
try it : y-api.bestvirtualgoods.com
You can always use DeepSeek v4 flash, mimo v2.5, hy3, and more models in the future for free.
try it : y-api.bestvirtualgoods.com
联系与支持 — Y-API
如何联系 Y-API 支持：邮件响应时间、我们能帮助的事项，以及常见问题的替代资源。
y-api.bestvirtualgoods.com
September 28, 2026 at 5:18 AM
Hello, this is Y-API, a large language model API that everyone can afford.
You can always use DeepSeek v4 flash, mimo v2.5, hy3, and more models in the future for free.
try it : y-api.bestvirtualgoods.com
You can always use DeepSeek v4 flash, mimo v2.5, hy3, and more models in the future for free.
try it : y-api.bestvirtualgoods.com
Who the FUCK is using Tencent Hy3? If you are such a person please tell me. I'm so confused.
June 11, 2026 at 3:56 PM
Who the FUCK is using Tencent Hy3? If you are such a person please tell me. I'm so confused.
🤖 Tencent unveils 295B-parameter Hy3 model for reasoning and agentic tasks
Tencent's release of the 295B parameter Hy3 model marks a shift towards large scale Mixture of Experts (MoE) models aimed at reasoning, agentic workflows,...
#DeepLearning #GenerativeAI #AIInference #AI #AIPulse
Tencent's release of the 295B parameter Hy3 model marks a shift towards large scale Mixture of Experts (MoE) models aimed at reasoning, agentic workflows,...
#DeepLearning #GenerativeAI #AIInference #AI #AIPulse
Read the full article →
www.synestesia.uk
July 7, 2026 at 9:38 AM
🤖 Tencent unveils 295B-parameter Hy3 model for reasoning and agentic tasks
Tencent's release of the 295B parameter Hy3 model marks a shift towards large scale Mixture of Experts (MoE) models aimed at reasoning, agentic workflows,...
#DeepLearning #GenerativeAI #AIInference #AI #AIPulse
Tencent's release of the 295B parameter Hy3 model marks a shift towards large scale Mixture of Experts (MoE) models aimed at reasoning, agentic workflows,...
#DeepLearning #GenerativeAI #AIInference #AI #AIPulse
Tencent releases Hy3-preview, its first AI model developed under former OpenAI researcher Yao Shunyu; the model features 295B parameters, down from HY2's 400B (Vincent Chow/South China Morning Post)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
April 23, 2026 at 10:15 AM
Tencent releases Hy3-preview, its first AI model developed under former OpenAI researcher Yao Shunyu; the model features 295B parameters, down from HY2's 400B (Vincent Chow/South China Morning Post)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
It's kind of amazing how cheap some of those tokens are too. The top two models, DSV4 Flash and Hy3 preview, are $0.10-0.14 and $0.06-0.07 per Mtok in respectively!
Cheap models make up a large share of OpenRouter tokens, but are a trivial share of revenue. Here's a table with the proportion of tokens and dollars below various blended prices. Models with a blended price <$0.50 make up 71% of tokens but only 10% of revenue.
June 8, 2026 at 4:30 PM
It's kind of amazing how cheap some of those tokens are too. The top two models, DSV4 Flash and Hy3 preview, are $0.10-0.14 and $0.06-0.07 per Mtok in respectively!