Hy3 is a Tencent-developed language model surfaced on the LLM Gateway through the NovitaAI route, presented alongside a parallel Tencent Cloud deployment. Independent coverage describes the underlying model as a very large system of roughly 295 billion parameters that uses sparse routing, activating only about 21 billion parameters during inference to balance capability and efficiency. That mixture-of-experts style architecture positions Hy3 as a heavyweight general-purpose text model offered to developers through an aggregator path rather than direct Tencent infrastructure, with calls on the sibling Tencent Cloud route reported to fall under Chinese jurisdiction.
In practical terms, Hy3 gained visibility by topping OpenRouter usage charts in mid-July 2026, suggesting strong adoption among developers experimenting with high-capacity open-style models. The report also highlighted a simple base-URL swap that lets the same model run free inside the Codex CLI, lowering the friction for hands-on evaluation. Because the model carries a 262K-token context window, it is well suited to long-document reasoning, code generation, and tool-assisted workflows where extensive prompt history or repository context needs to be carried into each turn. Teams looking for a large, sparsely routed text model accessible through a unified gateway will find Hy3 a notable option in this tier.
