Skip navigation
Tencent has just released a new model featuring 224 billion (224B) parameters, with 12 billion (12B)…
0:00 / 0:59

Tencent has just released a new model featuring 224 billion (224B) parameters, with 12 billion (12B)…

Sean Wallace

@sean_wallace_50K followers

Visit on X
154 likesDownloadOriginal post

69,423 views

Tencent has just released a new model featuring 224 billion (224B) parameters, with 12 billion (12B) active at any given time. It's priced at just $0.07 per million (1M) input tokens. Even more remarkably, it slightly outperforms Tencent’s own 770-billion (770B) parameter model, the Hy4 preview version, while costing only a fraction of the price. How is this possible? Let’s take a closer look. 🧵 Tencent Hy WorkBuddy Use WorkBuddy for free until November 30. Give it a try at no cost!

29 Comments

Comments from the original post on X (Twitter)

  • Kiage Clinton, KC's profile picture

    @iam_ochora_3

    One test won’t settle everything, but it’s more concrete than a leaderboard.

  • 🆁🅰🅽🅹🅾🆂's profile picture

    @WhoseKeli

    Keeping the prompt unchanged gives people a clearer starting point for comparison.

  • GOATFATHER🇰🇪's profile picture

    @ClintonioKE

    A slightly better score is interesting. A usable result at lower cost is more relevant to my work.

  • Stones's profile picture

    @josephstoness

    One expensive retry can change an otherwise attractive cost comparison.

  • Jayden Cole's profile picture

    @Jjay_Cole

    Lower input prices appeal, but completed runs determine actual spending.

  • Sean Wallace's profile picture

    @sean_wallace_

    For a proper comparision, I used the same prompt in Hy4 Model Hy4 Flash: 👉 AI Credits: 84$ 👉 time: 46 min Hy4 Model took: 👉 AI Credits: 130$ 👉 time: 63 min And here's the output 👇

  • Betty's profile picture

    @worldofBettty

    The comparison improves when you inspect what each model actually delivered.

  • RICHIE..😎's profile picture

    @RichiyChris

    Mentioning active parameter counts alongside headline figures makes comparisons clearer.

  • Lionel's profile picture

    @Lionelsecond

    Results should be judged against the prompt, not competitors alone.

  • Tyla's profile picture

    @Tylahworld

    A follow up edit would reveal how well each model maintains context.

  • Linn's profile picture

    @ItsLindajames

    Efficiency deserves attention without treating model size as everything.

  • Abuga with a G🐐's profile picture

    @abuga_01

    The same-prompt comparison is more useful to me than the parameter count.

  • Hope🇰🇪's profile picture

    @kijananoma

    The difference matters more when you’re running the task repeatedly.

  • Fatuma's profile picture

    @ItsFatumah

    Testing everyday tasks would make an interesting and practical follow up.

  • George's profile picture

    @Georgiekennn

    The quicker option suits early drafts when output quality remains consistent.

  • Bonita🤎's profile picture

    @bonita_254

    A few revision rounds would make a useful follow-up to this comparison.

  • E's profile picture

    @I_am_mutoro

    Finishing sooner also means getting to review the result sooner.

  • CATHEY's profile picture

    @Catheyloyal

    Launch figures encourage trying it, but output quality encourages staying.

  • ETHAN JAMES's profile picture

    @EthanisJames

    Recording failed attempts would make this comparison even more informative.

  • Talanta's profile picture

    @Talanta_boy

    Glad the comparison includes an actual task rather than just the launch figures.