Synthszr Charts — die großen AI-Marken im Wettkampf ums Podium
synthszr charts
tencent

Hy3

#147 in Frontier-Spraakmodelle

tencent · v3 · siet 2026-04-23 · 10× · tolest 30. Juni 2026

7
Momentum

Tencent Hy3 preview is an open-source language model with a Mixture-of-Experts (MoE) architecture, released and open-sourced on April 23, 2026. It features 295 billion total parameters with 21 billion active parameters per forward pass, plus 3.8 billion MTP layer parameters. The model supports a 256K token context window and offers three configurable reasoning modes (disabled, low, high). It is the first model trained on Tencent's rebuilt pre-training and reinforcement learning infrastructure and is available on Tencent Cloud TokenHub, OpenRouter, Hugging Face, ModelScope, and GitCode.

Momentum-Verloop
04.04.03.07.

Features

Key Benchmark (%)SWE-bench Verified: 74.4% | Terminal-Bench 2.0: 54.4% | BrowseComp: 67.1% | WideSearch: 70.2% (all reported by Tencent Hy Team)
Context Window (Tokens)256,000 tokens (256K); listed as 262,144 tokens at OpenRouter/API level
LicenseTencent Hy Community License Agreement (custom; includes usage restrictions above >100M MAU and limitations on using model outputs to train other AI models)
MultimodalityText-only (no native image/audio/video input); supports tool use/function calling and structured outputs
PlatformTencent Cloud TokenHub (API), OpenRouter, Hugging Face (tencent/Hy3-preview), ModelScope, GitCode; self-hosting via vLLM and SGLang
Price per 1M TokensInput: $0.063/M tokens; Output: $0.21/M tokens (OpenRouter pricing). Tencent Cloud TokenHub: CNY 1.2/M input, CNY 4/M output (approx. $0.17-0.18 / $0.59)
Release DateApril 23, 2026

Belege (10)

Mehr Produkten in disse Kategorie: Frontier-Spraakmodelle

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.