Tencent Launches Hy4 Preview Open-Weight AI Model at Low Cost

Tencent has launched Hy4 Preview, a low-cost, open-weight AI model designed to compete with leading systems from OpenAI, Anthropic, DeepSeek, Zhipu AI, and Moonshot AI.

Unlike proprietary models such as GPT-5.6 and Claude Opus 5, Hy4 Preview offers publicly available model weights that developers can download and deploy independently. It also comes with relatively affordable API pricing, starting at $0.042 per million cached input tokens, while regular input costs $0.834 and output costs $2.501 per million tokens.

What Are the Core Specifications of Hy4 Preview?

The model has 770 billion total parameters, activates 49 billion parameters per token, and supports a one-million-token context window. Tencent says it delivers significant improvements in software engineering, office productivity, game development, and scientific research.

Hy4 Preview uses a Mixture-of-Experts architecture, which activates only part of the model's total parameters for each token. The model contains 78 layers, including 77 MoE layers with 256 routed experts and one shared expert per layer. It also uses Gated DeepSeek Sparse Attention and identity Hyper-Connections to improve information processing.

Tencent says it increased model size, context length, and training data while expanding both pre-training and post-training. The company describes Hy4 Preview as its largest generation-over-generation improvement so far.

How Does Hy4 Preview Compare to Major AI Models?

According to Tencent's published benchmarks, Hy4 Preview outperforms several competing frontier models in selected coding and productivity tests. However, Claude Opus 5, GPT-5.6 Sol, and other rivals remain ahead in some areas.

Tencent evaluated Hy4 Preview against several leading AI models using software engineering, agent-based, and reasoning benchmarks. The comparisons include Claude Opus 5, GPT-5.6 Sol, DeepSeek V4 Pro, GLM 5.3, and Kimi K3.

Hy4 achieved 82.9 on SWE-bench Multilingual, exceeding the listed scores for DeepSeek V4 Pro, GLM 5.3, Kimi K3, and GPT-5.6 Sol. However, Claude Opus 5 performed better on that test. Hy4 also achieved a score of 64.0 on SWE Atlas Codebase Q&A, outperforming all five competitors shown in the evaluation table.

What Improvements Does Hy4 Offer Over Hy3?

Tencent also compared its new model with the previous Hy3 generation, showing notable improvements in coding, tool use, research, and long-context processing. One of the largest improvements appeared on DeepSWE, where the score increased from 28.0 to 64.3.

The model also improved from 30.8 to 64.0 on SWE Atlas Codebase Q&A and from 3.0 to 17.5 on ProgramBench. Tencent says these gains resulted from larger-scale training and greater use of real-world engineering tasks.

The model also performed competitively on benchmarks designed to measure its ability to search information, work with documents, and use external tools. On JobBench, its score rose from 34.6 to 61.7, while its OneMillionBench result improved from 51.5 to 65.4.

How Did Hy4 Perform in Blind Evaluations Against Competitors?

Tencent conducted a separate blind evaluation involving 163 internal experts and 203 software engineering tasks, assessing outputs produced by Hy4 Preview, GLM 5.3, and Kimi K3. Hy4 received an average score of 2.99 out of 4, compared with 2.92 for GLM 5.3 and 2.94 for Kimi K3.

Hy4 performed slightly better in both comparisons, although the differences in average scores were small. The tests were conducted internally by Tencent, so they should not be treated as independent confirmation of overall model superiority.

What Practical Capabilities Were Demonstrated for Hy4?

Tencent developed Hy4 Preview with input from software engineers, game developers, financial analysts, and security specialists to create training data focused on practical tasks. In software engineering, Tencent says Hy4 can handle longer development tasks while improving front-end design and interaction quality.

For office work, Tencent demonstrated a workflow involving 24 reimbursement claims, requiring the model to examine several policy versions, staff records, budgets, invoices, and emails. In game development, Hy4 created a cartoon adventure game featuring a penguin exploring a snowfield, collecting berries, opening chests, and fighting monsters.

Tencent also reported improvements in scientific research, particularly in AI research, molecular dynamics, condensed matter physics, and pure mathematics. In one experiment, Hy4 managed several Codex sessions simultaneously while working on a small-model post-training task.

Where Can Developers Access Hy4 Preview?

Released on August 28, 2026, Hy4 Preview is available through Tencent Cloud, OpenRouter, and Tencent's AI applications, while developers can also download model weights through supported open-source repositories. Tencent provides both the standard Hy4 Preview model and an FP8 version under the Apache 2.0 license.

Despite benchmark improvements, Tencent says the model still has problems that require further development, including spending too much time reasoning through complicated tasks and repeatedly checking its own work. Tencent expects further improvements through additional training and user feedback.