Hy3 Preview

Hy3 Preview appears as a Tencent-branded language model entry on public model comparison trackers, where it is grouped with Hy3 in a same-family comparison pair. This pairing is notable because it situates Hy3 Preview within a lineage that practitioners familiar with Hy3 can map against, even though the supplied sources do not expose parameter counts, training data scale, or architectural lineage details for the preview variant. The preview framing suggests the model is positioned as an intermediate or evaluation-ready release rather than a finalized production configuration.

From a practical standpoint, the model is documented on comparison platforms that surface it alongside its sibling, giving teams a familiar reference point when weighing migration or evaluation paths. The same-family treatment means that documented cost, context, and runtime differences can be used as practical selection signals once those values are verified, but public scores carry uncertainty and shared benchmark results are not available in the current evidence base. Teams evaluating Hy3 Preview should therefore treat the comparison framing as a structural signal of intent rather than a confirmed performance verdict, and rely on their own workload testing before committing to deployment.

AIHubMixhy3-previewHy

Open provider page Provider docs

Quick Info

Provider: AIHubMix
Model key: hy3-preview
Release date: Apr 20, 2026
Last updated: Apr 20, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.17
Output token cost: $0.566661

Limits

Output tokens: 128,000 tokens
Context window: 256,000 tokens

Transparent token rates

Compare Hy3 Preview pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Hy3 Preview

AIHubMix

CoveragePreview

Tencent's Hy3 preview is a 295B-parameter Mixture-of-Experts model with 21B activated parameters and a 3.8B MTP layer, developed by the Tencent Hy Team and described on its official Hugging Face model card as the first model trained on the team's rebuilt infrastructure. The card lists 80 layers (excluding the MTP layer The same model card highlights four capability areas: STEM and reasoning gains on FrontierScience-Olympiad, IMOAnswerBench, the Tsinghua Qiuzhen College Math PhD qualifying exam (Spring '26), and the China High School Biology Olympiad (CHSBO 2025); context learning and instruction following measured on Tencent's own CL

AIHubMix

CoveragePreview

Tencent Cloud's techpedia entry, titled "Hy3 preview: The First Step of Tencent Hunyuan's Rebuild" and dated July 27, 2026, recaps the April 23, 2026 open-source release of Hy3 preview as a hybrid fast-and-slow-thinking Mixture-of-Experts language model with 295B total parameters, 21B active parameters, and a 256K cont The entry positions Hy3 preview as Hunyuan's most intelligent model to date with major gains in reasoning, instruction following, in-context learning, code, and agent capabilities. It frames the April release as the first step of a broader Hunyuan rebuild rather than a standalone product launch, and links onward deploy

AIHubMix

CoveragePreview

The Baidu encyclopedia entry confirms that Hy3 Preview is an artificial intelligence model released and open-sourced by Tencent's Hunyuan team on April 23, 2026, following its reorganization. The model adopts a Mixture of Experts architecture with 295 billion total parameters and 21 billion activated parameters, suppor The same entry reports that Hy3 Preview's token usage reached ten times that of the previous-generation Hy2 model, with significant usage in code and Agent-type scenarios, and enhanced capabilities in complex reasoning, instruction following, code, and Agent tasks. An April 29, 2026 OpenRouter global large model API ca