Terminal Bench 4.0 从 preview 的 26.3 涨到 44.4，和闭源头部比还差一截：
This is a AI post classified by AGK Intelligence as Frontier models (a model release), kept by the AI Radar because it carries real work, not commentary.
腾讯的 Hy4 正式版出了，加了看图和电脑操作。 Terminal Bench 4.0 从 preview 的 26.3 涨到 44.4，和闭源头部比还差一截： 1. Claude Opus 5.5 66.4 2. GPT-6.1 Sol 55.5 3. Hy4 44.4 最接近的是电脑操作，OSWorld 2.1 上 Hy4 78.9，Opus 5.5 81.8。分数都是腾讯自己给的。 权重还没放，说是这几天和 system card 一起出，vLLM 和 SGLang 在做适配。更便宜更快的 Hy4 Flash 也说很快会出。 8 月的 preview 是 770B 总参数、49B 激活。正式版要还是这个规模，4-bit 也要 380GB 以上，我两台 DGX Spark 加起来 256GB 装不下，能在家跑的只能等 Flash。
Posted by Wei (7.1k followers) 1 days ago · 3 likes · 447 views · view the original post on X. Kept by the AI Radar as Frontier models.
More AI work like this
- Qwen 4 vs Deepseek Flash one shot Christmas Village. — @loktar00
- Gemini 4 Carbon just took ANOTHER #1 spot!!! — @StatsWire
- Dwarkesh Patel says OpenAI's newest models now beat Anthropic's at a Mafia style game… — @CEOinterview
- We’re launching VIALS-AA, a leaderboard for Handshake AI’s VIALS benchmark for visual… — @ArtificialAnlys
- Day 0: K3.1 Pro https://inferencex.semianalysis.com/inference/kimi-k31-pro — @AnushElangovan
- 🚨 Gemini 4 Hydrogen new checkpoint output! — @pankajkumar_dev
- Qwen4 just did Minecraft in space. 🛰️👀 — @aijoey
- Gemini 4 Hydrogen is peak flash model 🔥🔥 — @HarshithLucky3
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.3k posts from 5k X accounts over the last 14 days, 10.4k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 17:49 UTC. Full method.