Tencent releases Hy4 preview with 770B parameters and 1M context
Tencent has released and open-sourced Hy4 preview, a 770-billion-parameter mixture-of-experts model with 49 billion active parameters and a context window above one million tokens. AIMI saw it reach OpenRouter and OpenCode Go later the same day.
Synthesized via Fish Audio S2.1 Pro in British English. Natural editorial summary, not verbatim reading.
Key Takeaways
- check_circleTencent announced and open-sourced Hy4 preview on 28 August 2026. The model has 770 billion total parameters, 49 billion active parameters and a context window above one million tokens.
- check_circleTencent says Hy4 preview is available through WorkBuddy, CodeBuddy, Yuanbao, ima, Tencent Cloud TokenHub and OpenRouter. The company offered free access on WorkBuddy and CodeBuddy for two weeks after launch.
- check_circleAIMI first saw tencent/hy4-preview on OpenRouter at 08:31:32 SAST, eight minutes and one second after Tencent Hy announced it. OpenCode Go exposed hy4-preview at 17:08:48 SAST.
- check_circleTencent reports a blind internal evaluation score of 2.99 out of 4.00, slightly ahead of GLM-5.3 and Kimi K3. That comparison has not been independently verified.
Tencent releases a preview built for practical work
Tencent Hy announced Hy4 preview at 08:23:31 SAST on 28 August 2026. The post described a 770-billion-parameter mixture-of-experts model with 49 billion active parameters and a one-million-token context. Tencent presented it as a model for coding, productivity and other real-world work, not as a finished general-availability product.
Tencent's corporate newsroom published its release notice at 10:48:45 SAST the same day. It says the model is open source and can be used through Tencent products, Tencent Cloud TokenHub and OpenRouter. The official Hugging Face and GitHub links provide the model card and release materials.
The provider route arrived eight minutes later
AIMI records OpenRouter provider metadata at 08:09:35 SAST, before Tencent's public X announcement. That timestamp describes when OpenRouter created its catalogue record. It is not the maker release time.
AIMI first saw openrouter/tencent/hy4-preview at 08:31:32 SAST. That is exactly 8 minutes and 1 second after Tencent Hy's announcement, and 22 minutes and 57 seconds after the OpenRouter provider metadata time. OpenRouter lists a 1,048,576-token context, 64,000 maximum output tokens, reasoning, tools, function calling and structured outputs. Its current price is $0.834 per million input tokens and $2.501 per million output tokens.
OpenCode Go added the same model later
OpenCode Go listed hy4-preview at 17:08:48 SAST, 8 hours, 37 minutes and 16 seconds after the OpenRouter endpoint observation. Its official documentation places Hy4 preview in the Go model list and exposes the route at https://opencode.ai/zen/go/v1/chat/completions.
OpenCode Go documents Hy4 preview at the same token prices as the OpenRouter page and assigns it a $30 usage allowance in the subscription table. The Go route snapshot does not expose the same capability metadata as OpenRouter, so developers should not assume that the two gateways behave identically.
What Tencent reports, and what remains untested
Tencent says an internal blind evaluation involving 163 experts and 203 engineering tasks gave Hy4 preview an average score of 2.99 out of 4.00. The company reports 2.92 for GLM-5.3 and 2.94 for Kimi K3 in the same comparison.
Those numbers come from Tencent's own evaluation. The release materials do not provide an independent replication, and the preview label signals that the model is still early. Treat the benchmark chart as useful context, not as a production guarantee. Test latency, tool calls, long-context retrieval and failure handling on the work your team actually runs.
Access through the AZ Labs AI Gateway
The exact OpenRouter route is tencent/hy4-preview. OpenCode Go uses hy4-preview on its own subscription endpoint. Tencent also names TokenHub and several first-party products as access points. These are different delivery paths with different authentication, limits, pricing and data policies.
AZ Labs customers can use the AI Gateway to keep the application integration separate from the provider choice, then compare Hy4 preview with other routes against their own tasks. Do not send confidential material to an early preview until you have checked the current provider terms and retention policy.
Visual Architecture & Benchmarks
Official maker scorecard graphics, diagrams, and benchmarks. Click any image to inspect in full resolution.
Frequently Asked Questions
Is Hy4 preview a real new model release?
Yes. Tencent Hy announced Hy4 preview on 28 August 2026, and Tencent's corporate newsroom published a same-day release notice describing it as open source.
What is the exact OpenRouter route?
The exact OpenRouter route is tencent/hy4-preview. OpenRouter lists it at $0.834 per million input tokens and $2.501 per million output tokens in the snapshot used for this article.
Where is Hy4 preview available?
Tencent names WorkBuddy, CodeBuddy, Yuanbao, ima, Tencent Cloud TokenHub and OpenRouter. OpenCode Go also lists hy4-preview on its official model endpoint.
Are Tencent's benchmark results independently verified?
No. The reported comparison comes from Tencent's internal blind evaluation. Test the model on your own workload before relying on the reported ranking.
Is the model suitable for confidential production work?
Check the current terms, retention policy and provider route before sending sensitive data. Hy4 preview is an early release, and its availability and limits may change.
Related Frontier Models & Releases
Explore verified specifications, benchmark results, and route pricing across alternative models in this class.
DeepSeek releases V4.1 Flash with native multimodal vision, 552B MoE and lower API rates
DeepSeek has launched DeepSeek-V4.1-Flash, a 552B-parameter mixture-of-experts model featuring a novel Causal Encoder–Decoder architecture with just 8B active input and 16B active output parameters. The release brings native vision, compresses KV cache storage by up to 8x, and slashes off-peak API pricing to $0.15 per million input tokens.
OpenAI Releases GPT-6 Astra: Next-Generation Flagship with 1.05M Context and Deep Multimodal Reasoning
OpenAI has officially launched GPT-6 Astra, its frontier flagship model featuring a 1,050,000-token context window, 128,000 max output tokens, and native tool-use for autonomous agent workflows.
Alibaba Qwen Releases Qwen3.8 Max: 2.4-Trillion Parameter Flagship with 1M Multimodal Context
Alibaba's Qwen team has launched Qwen3.8 Max, a 2.4T parameter Mixture-of-Experts model offering 1M token context, native video perception, and deep agent tool orchestration.
Primary Sources
- Tencent corporate Hy4 preview release
- Tencent Hy official announcement on X
- Tencent Hy4 preview research page
- Tencent Hy4 preview model card
- Tencent Hy4 preview GitHub repository
- OpenRouter Hy4 preview route
- OpenCode Go documentation
- OpenCode Go models endpoint
- Timestamped AZ Labs AIMI release evidence
- AZ Labs AIMI model index methodology
- Route models through the AZ Labs AI Gateway