---
title: "Tencent releases Hy4 flash: 224B parameters at 2 yuan per million output tokens"
date: 2026-11-19
category: Foundation Models
site: NeuroAI
canonical: https://neuroai.site/a/na-auto-2026-11-19-tencent-hy4-flash
language: en
---

# Tencent releases Hy4 flash: 224B parameters at 2 yuan per million output tokens

> Tencent's Hy4 flash has 224B total parameters with 12B active, priced at 2 yuan per million output tokens, with headline results reported on Tencent's internal WorkIndex benchmark.

Tencent has released Hy4 flash, a mixture-of-experts model with 224 billion total parameters and 12 billion active, priced at 2 yuan per million output tokens.

The architecture explains the price. Sparse activation means each token touches only a fraction of the weights, so inference cost tracks active parameters rather than total size - the same pattern that made open-weight Chinese models competitive on price through the past year.

The company also published standard, FP8 and base checkpoints, with documented download access, licence terms and deployment conditions. Releasing weights alongside an API is now a standard playbook: it seeds enterprise on-premise deployments while keeping the hosted service as the default.

One caveat for anyone benchmarking: the headline results are reported on Tencent's internal WorkIndex evaluation, so external comparison on public suites is still pending.

At 2 yuan per million output tokens, the model lands in the band where agentic workloads - many calls, long outputs - become economically plausible.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-auto-2026-11-19-tencent-hy4-flash
Free to quote with attribution and a link to the original.
