Tencent: Hy3

Text input Text output
Author's Description

Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort:...

Key Specifications
Cost
$$
Context
202K
Parameters
295B (Rumoured)
Released
Jul 06, 2026
Speed
Ability
Reliability
Supported Parameters

This model supports the following parameters:

Stop Max Tokens Structured Outputs Seed Reasoning Top P Frequency Penalty Presence Penalty Temperature Include Reasoning Logit Bias Tools Tool Choice Min P
Features

This model supports the following features:

Tools Structured Outputs Reasoning
Performance Summary

Tencent's Hy3, a 295B-parameter Mixture-of-Experts model (21B active, 192 experts with top-8 routing) designed for reasoning, agentic workflows, and production use, demonstrates a balanced performance profile. With a context length of 202752, Hy3 exhibits competitive response times, performing among the 46th percentile across various benchmarks. It also offers cost-effective solutions, ranking in the 67th percentile for pricing. A standout feature is its exceptional reliability, boasting a 99% success rate, indicating consistent and evaluable responses with minimal technical failures. In specific benchmark categories, Hy3 shows strong capabilities in hallucination prevention, achieving 98.0% accuracy by appropriately acknowledging uncertainty, placing it in the 69th percentile. Its instruction following is robust, with 74.0% accuracy across complex multi-layered instructions, ranking in the 75th percentile. The model also performs well in email classification, reaching 98.0% accuracy (65th percentile). While its accuracy scores are generally high, the duration for some tasks, such as hallucination and instruction following, falls into the lower 35th-36th percentile, suggesting areas for potential speed optimization despite its overall competitive speed ranking. Overall, Hy3 is a reliable and capable model, particularly strong in complex reasoning and instruction adherence.

Model Pricing

Current Pricing

Feature Price (per 1M tokens)
Prompt $0.14
Completion $0.58
Input Cache Read $0.035

Price History

Available Endpoints
Provider Endpoint Name Context Length Pricing (Input) Pricing (Output)
AtlasCloud
AtlasCloud | tencent/hy3-20260706 202K $0.14 / 1M tokens $0.58 / 1M tokens
Benchmark Results
Benchmark Category Reasoning Strategy Free Executions Accuracy Cost Duration
Other Models by tencent