Hy4 by Tencent: model card
What it is
Hy4 is a agentic model from Tencent. It ranks #14 of 59 on its Arena board, which puts it in the upper half. The model runs multi-step tasks in an environment: reads files, runs commands, uses tools.
Key numbers: net improvement 5.75%; sessions 21 617; licence Proprietary.
What this card does not contain: our own test runs, latency measurements, user reviews or taste judgements. Everything stated here comes from the public Arena.ai leaderboard and specs published by the developer. A missing parameter is marked as not listed rather than guessed.
Leaderboard numbers
- Rank: #14 of 59
- Net improvement: 5.75%
- Sessions: 21617
- Licence: Proprietary
Who it is for
- Teams choosing a default model for a product and wanting the whole board, not just the top row.
- Anyone comparing price per token with rank, looking for the point where a cheaper model stops failing.
How to read these numbers
An Arena score comes from blind pairwise comparisons, not from an absolute quality scale. A few points between neighbouring rows is often inside the error bars, so the top of a board is better read as ‘roughly one tier’. Price, context and licence are stated specs and do not feed the score.
What to check yourself
- Whether the model is available in your region and whether access terms changed since the snapshot.
- Whether you really need the stated context: attention quality on long input is often lower.
- How price behaves on your prompt mix: the input/output ratio can matter more than the headline figure.
Source
Leaderboard snapshot: 11 November 2026.
_Prepared with AI assistance from the public Arena.ai leaderboard. This is not first-hand experience with the model and not an independent review._
Коментарі (1)
Увійдіть, щоб залишити коментар
Авторизуйтесь через Google або свій обліковий запис. Ви зможете коментувати повністю анонімно 🕵️, увімкнувши галочку анонімності.
Увійти / ЗареєструватисьReview notes and sources: Hy4 (Tencent, #14 of 59).
What to know before putting this model to work:
Checked links:
_Written by a listing.ee AI persona from open sources. Not a user review and not the result of independent testing; leaderboard numbers are a snapshot taken 11 November 2026._