Hy3

Hy3 is presented as a large sparse language model built around a Mixture-of-Experts design, with 295 billion total parameters and 21 billion active for each token. That configuration is intended to preserve substantial model capacity while limiting the computation involved in inference, making it a candidate for organizations that need capable generation without activating the entire network on every request.

The available reporting also positions Hy3 as a Tencent release with a 256,000-token context window. Its scale and long-context design make it most relevant to substantial text-generation workloads, complex reasoning over extended material, and applications that can accommodate a larger sparse model. No supplied evidence supports benchmark leadership, specific training methods, or measurable advantages over competing models, so performance should be evaluated against the target workload.

OpenCode Gohy3Hy

Open provider page Provider docs

Quick Info

Provider: OpenCode Go
Model key: hy3
Release date: Jul 6, 2026
Last updated: Jul 6, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.14
Output token cost: $0.58

Limits

Input tokens: 192,000 tokens
Output tokens: 128,000 tokens
Context window: 256,000 tokens

Transparent token rates

Compare Hy3 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

OpenCode Go

Official sourceDocumentation

OpenCode Go is a $10/month subscription offered by OpenCode that bundles vetted open-weight coding models behind a single API key, sitting alongside other providers in the OpenCode ecosystem. According to the provider documentation, subscribers sign in to OpenCode Zen, copy their API key, then run the /connect command For developers, the practical impact is that Hy3 can be reached through OpenCode Go's unified endpoint without negotiating separate provider agreements, which is relevant given Tencent's own positioning of Hy3 as a coding- and agent-oriented model with a 295B-total / 21B-active MoE architecture and a 256K context windo

OpenCode Go

CoveragePreview

The Hugging Face model card for tencent/Hy3-preview documents the technical architecture of the Hy3 model developed by the Tencent Hy Team. It is a Mixture-of-Experts model with 295 billion total parameters, 21 billion activated parameters, and an additional 3.8 billion MTP layer parameters, built on 80 layers with 64 The model card also highlights benchmark results across STEM and reasoning (FrontierScience-Olympiad, IMOAnswerBench, Tsinghua Qiuzhen College Math PhD qualifying exam Spring '26, China High School Biology Olympiad 2025), context learning and instruction following using Tencent-built CL-bench and CL-bench-Life evaluati

OpenCode Go

Coverage

Tencent's official press release of August 5, 2026 announces broader international access to Tencent Hy3 (Tencent Hy, formerly known as Tencent Hunyuan) following the model's official release on July 6. The announcement describes Hy3 as a hybrid fast-and-slow-thinking Mixture-of-Experts model with 295 billion total par The press release also reports that Hy3 recorded more than 68 times as many API calls as its previous-generation model and ranked first globally on OpenRouter's usage leaderboard within one week of launch, and lists developer-platform integrations including Hermes, Kilo, Cline, OpenClaw, OpenCode, and Cherry Studio, al

OpenCode Go

CoverageRelease Notes

Tencent announced global availability of Hy3 on August 5, 2026, following the model's official release on July 6, 2026, broadening access for developers, end users, and enterprises beyond China. Hy3 retains the hybrid fast-and-slow-thinking Mixture-of-Experts architecture with 295 billion total parameters, 21 billion a Distribution channels named in the announcement include WorkBuddy (free to users worldwide until August 31, 2026 Pacific Time), Tencent Design Miora, and Tencent Cloud TokenHub, with developers able to connect via API and integrate with coding extensions and third-party platforms. The release explicitly lists "OpenCode

OpenCode Go

CoverageBenchmark

The llm-stats.com model page, dated September 17, 2026, places Hy3 at an overall composite rank of 41 in a B tier, with capability rankings of 12 of 90 in Search and 14 of 119 in Long Context. The page lists Hy3 at a blended price of $0.16 per million tokens, positioning it between GPT OSS 120B and DeepSeek-V4.1-Flash The aggregator records Hy3 benchmark scores including USAMO 2026 at 30.24 of 42 (rank 3), DeepSearchQA at 0.91 (rank 5), GPQA Diamond at 0.90 (rank 25), and IMO-AnswerBench at 0.90 (rank 3), all sourced from Hugging Face and flagged as provider-reported. It also shows Math and Coding rankings of 33 of 327 and 35 of 267