Hy4 Flash drops free on Nous Portal
Tencent Hunyuan's new Hy4 Flash model is available for free on Nous Portal for a two-week promotional window. The 224B-parameter MoE model activates 12B parameters per token, targeting high-volume agentic tasks and document workflows with a 1-million-token context window.
Tencent is aggressively pushing into low-cost frontier inference, and partnering with Nous Research gives Hy4 Flash immediate credibility in the developer and agent ecosystems. Offering two weeks of free access is a classic developer land-grab to seed workflow integrations before paid API billing begins.
- –Extreme MoE efficiency: With 224B total parameters and only 12B active per token, Hy4 Flash achieves over 150 tokens per second while cutting KV cache overhead to 1/24th of Hy3.
- –Long-context utility: A native 1-million-token context window paired with rock-bottom pricing ($0.07/M input, $0.28/M output) makes it an aggressive competitor to Gemini 1.5 Flash and Claude 3.5 Haiku for document-heavy pipelines.
- –Nous Portal synergy: Placing the model directly inside Nous Portal lets developers immediately test it within the Hermes Agent ecosystem and MCP tooling without managing API keys or infrastructure.
- –Computer-use focus: The model is specifically fine-tuned for high-volume office tasks, slide deck generation, spreadsheet repair, and GUI/CLI automation, signaling Tencent's enterprise workflow ambitions.
DISCOVERED
1h ago
2026-11-16
PUBLISHED
1h ago
2026-11-16
RELEVANCE
AUTHOR
NousResearch