Open-weight reasoning and coding model · Official-source review 2026-10-09
Tencent Hy4 preview is a Mixture-of-Experts model with downloadable instruct and FP8 weights. The official repository documents 770 billion total parameters, 49 billion activated per token, a one-million-token context and developer deployment recipes.
For model-infrastructure teams evaluating long-context coding and research workloads with their own deployment.
Preview release has known excessive reasoning and over-verification behavior. Example serving recipe uses eight-way tensor parallelism; no hardware affordability or performance benchmark independently verified. Tencent’s internal evaluations are provider research, not our tests. Downloadable weights do not include free cloud compute or paid hosting. No model downloaded, deployed or invoked.
Review the repository and license, budget deployment hardware, then benchmark a scoped workload before adopting the preview.
Preview release has known excessive reasoning and over-verification behavior. Example serving recipe uses eight-way tensor parallelism; no hardware affordability or performance benchmark independently verified. Tencent’s internal evaluations are provider research, not our tests. Downloadable weights do not include free cloud compute or paid hosting. No model downloaded, deployed or invoked.
Official product evidence reviewed on 2026-10-09; product artwork inspected visually. No hands-on vendor application test was performed. Application performance, security claims and plan enforcement were not independently tested.
What it can do
Unknown is different from unavailable. Each fact carries its own evidence.
| Capability | Value | Evidence | Checked |
|---|---|---|---|
| Product workflow | Tencent Hy4 preview is a Mixture-of-Experts model with downloadable instruct and FP8 weights. The official repository documents 770 billion total parameters, 49 billion activated per token, a one-million-token context and developer deployment recipes. | Facts sourced | 2026-10-09 |
| Pricing and restrictions | Preview release has known excessive reasoning and over-verification behavior. Example serving recipe uses eight-way tensor parallelism; no hardware affordability or performance benchmark independently verified. Tencent’s internal evaluations are provider research, not our tests. Downloadable weights do not include free cloud compute or paid hosting. No model downloaded, deployed or invoked. | Facts sourced | 2026-10-09 |
Understand the total cost
Free
Free
Model released under Apache 2.0; running the weights requires substantial infrastructure. Hosted API or consumer-app pricing unconfirmed.
RunInfraRunInfra provides hosted open models through OpenAI- and Anthropic-compatible APIs. It offers prepaid token billing and workspace coding plans, with session-aware caching and separately reported cached-token usage.
Explore tool
The AI PlatformThe AI Platform combines people, model-backed specialists and project context in a desktop workspace. Embedded code/browser tools, shared channels and higher-tier workflows let teams review work and spending in one place.
Explore tool
chat-recallchat-recall stores coding conversations for full-text and knowledge-graph retrieval. It offers MCP access, analytics and code-related findings, with hosted and self-hosted deployment options.
Explore toolThe practical questions
No. This listing is based on dated official-source evidence and a visual artwork check. Unconfirmed costs and restrictions are identified explicitly.
Tencent Hy4 preview is a Mixture-of-Experts model with downloadable instruct and FP8 weights. The official repository documents 770 billion total parameters, 49 billion activated per token, a one-million-token context and developer deployment recipes.
Model released under Apache 2.0; running the weights requires substantial infrastructure. Hosted API or consumer-app pricing unconfirmed.. This record lists ongoing free access; check the plan limits before starting.
No paid monthly price is listed; this record treats the product as free to start. See the plan cards for entitlements, billing commitments and seat minimums.
Official recipes expose an OpenAI-compatible chat API through vLLM or SGLang; local deployment and infrastructure are required.. API access and subscription access may have different terms; consult the linked sources.
Preview release has known excessive reasoning and over-verification behavior. Example serving recipe uses eight-way tensor parallelism; no hardware affordability or performance benchmark independently verified. Tencent’s internal evaluations are provider research, not our tests. Downloadable weights do not include free cloud compute or paid hosting. No model downloaded, deployed or invoked.
No retained pricing changes yet. A current price alone does not establish a historical trend.
Current official product source read individually; product artwork inspected visually.
Read original source ↗Official product page reviewed for disclosed costs; unverified prices and allowances are identified in the listing.
Read original source ↗Facts apply to the named version and check date. Send a sourced correction if something changed.