Tencent just pulled back the curtain on Hy4 Preview — its next-generation large language model that brings serious firepower to the open-source arena. Announced on August 28, the model boasts 770 billion total parameters, with 49 billion activated parameters and a context window exceeding 1 million tokens.
That's not just a number. A 1M-token context means Hy4 can process entire books, lengthy codebases, or hours of video transcripts in a single pass — something most models still struggle with.
The Specs That Matter
Here's what makes Hy4 Preview noteworthy:
- 770B total parameters — placing it in the same league as DeepSeek V3 and Alibaba's Qwen-Max
- 49B activated parameters — using mixture-of-experts architecture for efficient inference
- 1M+ token context — among the longest available in any open model
- Open-source preview — weights available for research and commercial use
The model is already live on Tencent's WorkBuddy and CodeBuddy platforms, as well as Yuanbao and ima. API access is available through Tencent Cloud TokenHub and OpenRouter, priced at $0.834 per million input tokens and $2.501 per million output tokens.
Benchmark Performance: How Does It Stack Up?
Tencent conducted internal blind evaluations involving 163 experts and 203 engineering tasks. The results:
- Hy4 Preview: 2.99 out of 4.0 average score
- GLM-5.3 (Zhipu): 2.92
- Kimi K3 (Moonshot): 2.94
While the margins are slim, Hy4 Preview edges ahead of China's other top open models. More impressively, Tencent claims the model helped optimize its own training and inference systems, increasing end-to-end throughput by 31.8% compared to baseline.
For context, this is the same tier of performance as models costing significantly more to run — a testament to Tencent's MoE (Mixture of Experts) optimization.
Free Access, Strategic Move
Tencent is offering free access to WorkBuddy and CodeBuddy for two weeks. This is a classic Chinese AI lab play: give developers hands-on time with the model, build habit and dependency, then convert to paid API usage.
The pricing is aggressive too. At $0.834/M input tokens, Hy4 Preview undercuts many Western alternatives while delivering comparable or superior performance on long-context tasks.
What This Means for the Open-Weight Movement
Hy4 Preview continues the pattern we've seen from DeepSeek, Alibaba, and Zhipu: Chinese labs are racing to release capable open models faster than Western counterparts can respond.
Unlike OpenAI and Anthropic, which keep their frontier models behind API paywalls, Tencent is making Hy4 Preview accessible to anyone. The implications are significant:
1. Developers get a viable alternative. For teams building applications that need long context (legal document analysis, codebase understanding, research papers), Hy4 Preview offers a domestic Chinese option that doesn't rely on US infrastructure.
2. The open-weight ecosystem deepens. Each new release from Chinese labs adds to the pool of models that developers can fine-tune, deploy privately, and build upon without vendor lock-in.
3. Price pressure mounts. When a 770B-parameter model with 1M context costs less than $1 per million input tokens, it forces everyone else to reconsider their pricing.
🔥 Hot Takes
1. Tencent is playing 4D chess while the West plays checkers. Free two-week access? Check. Competitive pricing? Check. Open weights? Check. While OpenAI charges premium rates for GPT-4o, Tencent is giving away comparable capability and asking only for loyalty. This is how you build ecosystem dominance.
2. 1M token context isn't a gimmick — it's the future. Most models still cap at 128K or 200K tokens. Hy4 Preview's 1M+ context means you can feed it entire codebases, legal documents, or research libraries and get coherent, detailed analysis. That's not incremental improvement; that's a paradigm shift in what's possible.
3. The benchmark war is real, and China is winning. Hy4 Preview's 2.99 score against GLM-5.3's 2.92 and Kimi K3's 2.94 shows Chinese labs are no longer chasing — they're leading. The gap between "good enough" and "best in class" has vanished.
Bottom line: Tencent's Hy4 Preview isn't just another model release. It's a statement that China's AI ecosystem can produce frontier-capable, open-weight models at competitive prices — and do it faster than anyone expected. The question is no longer whether Chinese AI can compete. It's whether Western labs can maintain their premium pricing when open alternatives deliver equal or better performance at a fraction of the cost.