Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort: a direct no-think mode by default, plus low and high chain-of-thought modes for complex math, coding, and multi-step problems. With a 256K context window, Hy3 targets long-horizon tasks, including improved coreference resolution, multi-turn constraint tracking, and stable tool-calling that generalizes across agent scaffoldings. Tencent positions it as a reliable, cost-effective option across coding, document processing, financial analysis, game development, and frontend design, with a strong emphasis on grounded, anti-hallucination behavior that answers when grounded and flags when evidence is missing rather than fabricating.
All providers = every upstream serving this model supports it. Some providers = depends on which upstream handles the request. Default = the value sent when you leave the parameter unset.
| Parameter | Providers | Default |
|---|---|---|
| frequency_penalty | Some providers | Not sent by default |
| include_reasoning | All providers | - |
| logit_bias | Some providers | - |
| max_completion_tokens | Some providers | - |
| max_tokens | All providers | - |
| min_p | Some providers | - |
| presence_penalty | Some providers | Not sent by default |
| reasoning | All providers | - |
| reasoning_effort | All providers | - |
| repetition_penalty | Some providers | Not sent by default |
| response_format | Some providers | - |
| seed | Some providers | - |
| stop | Some providers | - |
| structured_outputs | Some providers | - |
| temperature | All providers | 0.9 |
| tool_choice | Some providers | - |
| tools | Some providers | - |
| top_k | Some providers | Not sent by default |
| top_p | Some providers | 1 |
All providers for this model are busy right now
Every upstream provider has hit its rate limit. The model comes back automatically once limits lift, usually within hours. Try again in a little while or switch to another model.
Request this model on Discord