Vibeleaderboard
← All Intel
Intel / post

HY 2.0 ships: 406B MoE, 256K context, SWE-bench Verified 53.0

Source
x.com
Date
Tencent Hy@TencentHunyuan

Tencent HY 2.0 is officially released. We are rolling out a major performance upgrade to our foundation model, now available via Tencent Cloud API. Built on a Mixture-of-Experts (MoE) architecture (406B total, 32B active parameters) and featuring a 256K context window, HY 2.0 delivers significant gains across key benchmarks. HY 2.0 Highlights: 🧠Reasoning: Achieved 73.4 on the IMO-AnswerBench (up nearly 20%), establishing top-tier math and scientific logic capabilities. 🛠️Coding & Agents: Explosive growth with SWE-bench Verified scoring 53.0 (from 6.0) and Tau2—Bench reaching 72.4 (from 17.1). ⚡Instruction Following: Enhanced stability and execution accuracy for complex constraints, producing more natural, less generic output. Two Optimized Versions:: HY 2.0 Think: Optimized for deep reasoning, code generation, and complex instruction following. HY 2.0 Instruct: Ideal for general chat, creative writing, and high-context multi-turn dialogue. Start building with HY 2.0 today. 🌐 Website: https://t.co/U3966JuQHi 🔗 API Access: https://t.co/BBNue9Awhy 📄 Documentation:

View on X

Context

Tencent's HY 2.0 announcement reports a substantial improvement on a software-repair test: its Think variant scores 53.0 on , compared with 6.0 for Hunyuan-T1-20250822 in the company's comparison chart. SWE-bench Verified measures the share of a human-filtered set of software issues that a system resolves. The reported gain therefore concerns completing tested repairs, rather than simply producing plausible-looking code. It remains a vendor-reported comparison under its setup.

Terms in this piece · Glossary
  • SWE-bench — The standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.
  • eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
  • mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
Read the source x.com
More from Tencent Hy
Recommended reads
Comments

Checking sign-in…

Loading comments…