What happened
Vercel announced that Hy4 from Tencent is available on AI Gateway. The model identifier is tencent/hy4. Vercel describes the model as supporting reasoning, tool calling, structured outputs, and image input. It lists a one-million-token context window, positioning that capacity for work that spans a codebase or a large collection of documents.
The announcement calls Hy4 the first model in the Hy line with vision. According to Vercel, it can identify fine detail in images and read complex charts. Vercel also says the model supports computer use in both command-line and graphical interfaces. Those features put several modalities and agent-oriented interfaces behind one model identifier, rather than limiting the release to text generation alone.
Vercel compares the release with Hy4 Preview. It says Hy4 follows instructions more closely and performs better with long context, long-running tasks, and deep research. The source does not include benchmark tables, task examples, methodology, or numerical results for those comparisons. The claims therefore describe the vendor’s positioning for the model rather than independently established performance differences.
Developers can call tencent/hy4 via Vercel’s AI SDK or its Chat Completions interface. For use in coding agents, Vercel instructs users to install the latest Vercel CLI and run npx vercel ai-gateway setup, then select the model in the agent. The announcement specifically names Claude Code, Codex, and Cursor among the supported destinations, and points users to a model playground for trying Hy4.
The access change includes promotional pricing: Hy4 is 50% off through December 31, 2026. The supplied announcement does not state the underlying input or output rates, the resulting discounted rates, or how the discount is calculated. Vercel says AI Gateway reflects provider pricing with no markup and charges no platform fee for inference, including Bring Your Own Key requests.
More broadly, Vercel presents AI Gateway as a layer for calling models, tracking usage and cost, and configuring retries, failover, and performance optimizations for higher-than-provider uptime. It also lists custom reporting, Zero Data Retention support, API-key budgets, and routing rules. None of those gateway features establishes Hy4-specific latency, quotas, regional coverage, or production reliability, but they describe the operational surface through which the new model is being offered.
What the archive adds
This is a new availability and pricing event, not a continuation of a documented Hy4 story. The archive has covered Exa’s hosted MCP search through AI Gateway, but that story concerns a search integration rather than Tencent’s model. It establishes prior coverage of the gateway as an integration surface, not prior coverage of Hy4.
Trend signal
This weakly supports agent economics moving to completed work. Hy4 is positioned for long-running work, tool use, and computer-use scenarios, while the temporary discount lowers the cost of evaluating it. For such workloads, a team needs to compare the cost and reliability of a completed task rather than just a model’s advertised capabilities. The source supplies no completion, reliability, latency, or cost-per-task measurements, so it does not show that Hy4 improves those economics in practice.
Why now
Vercel gives the promotion a stated end date: the 50% discount runs through December 31, 2026. That creates a defined evaluation window for teams already using AI Gateway. The announcement also provides a direct setup path for supported coding agents, so an existing gateway user can test the model without first building a standalone provider connection.
Why it matters
For a team already routing requests through AI Gateway, Hy4 is an additional model option that can be evaluated within an existing access and observability layer. The source describes a combination of image input, tool calling, structured outputs, and a one-million-token context window. That makes it relevant to workflows that combine code or lengthy documents with visual material, and to agent workflows that must return more constrained data than free-form prose.
The stated support for Claude Code, Codex, and Cursor matters because it reduces the operational distance between model availability and a practical trial. A team can select tencent/hy4 after running Vercel’s setup command rather than redesigning its calling layer. The discount reduces the immediate price of that experiment, although the announcement does not give enough rate information to calculate expected savings for a particular workload.
The useful decision is not whether a million-token context window or computer-use support sounds attractive. It is whether the model completes representative work reliably enough to earn a place in the stack. Teams should test their own tasks, including output quality, latency, tool-use success, structured-output validity, and total task cost. The announcement provides a capability list and vendor comparisons, not a guarantee that every named agent workflow behaves the same way or that the model is the best choice for a given task.
Limits and uncertainty
The source is Vercel’s product announcement. It does not disclose Hy4’s underlying pricing, the methodology behind claims versus Hy4 Preview, independent benchmark results, latency, uptime, throughput, or usage limits. It also does not explain how computer use is exposed in each supported coding agent, document regional availability, or specify whether every account can use the promotional offer.
The end date for the 50% discount is stated, but the source does not describe all eligibility conditions or how the promotion interacts with Bring Your Own Key. Vercel’s descriptions of vision, long-context behavior, instruction following, and deeper research performance should be treated as product claims until tested on the intended workflow. Gateway-level retries and failover do not themselves establish Hy4’s service quality.
Skill to build now
Create a small evaluation set from work your team already performs: one long-document or codebase task, one image or chart interpretation task if relevant, and one tool-calling or structured-output task. Run the same set through the current production model and tencent/hy4 via AI Gateway. Record task completion, correction time, latency, token usage, and failures rather than comparing only the raw response text. This turns the temporary discount into a bounded buying decision.
Reader takeaway
Hy4 is now an option in Vercel AI Gateway, with multimodal and agent-oriented capabilities described by Vercel and a 50% promotional discount through December 31, 2026. Existing Gateway users can try it as tencent/hy4 through supported coding-agent setup or the API. Evaluate it on real completed tasks, because the announcement does not supply the independent evidence needed to infer production value.