Tencent launches Hy4 flash multimodal model
Tencent has introduced Hy4 flash, a new multimodal AI model designed to balance speed and efficiency for daily operations like document analysis, research, and computer-assisted tasks. Featuring a 1-million-token context window and a Mixture-of-Experts (MoE) architecture with 224B total parameters (12B active), the model delivers decoding speeds exceeding 150 tokens per second. It is competitively priced at $0.07 per million input tokens and $0.28 per million output tokens, positioning it as an accessible option for high-volume workloads, with Tencent also planning open-weight versions for developers.
Tencent is aggressively positioning Hy4 flash to compete in the high-efficiency tier, targeting users who need reliable performance for routine operations without the premium cost associated with larger frontier models.
- –The combination of a 1M-token context window and low inference costs makes it highly attractive for high-volume document and data-heavy tasks.
- –The MoE architecture is critical to achieving a balance between large-scale knowledge (224B parameters) and fast, efficient active inference (12B active parameters).
- –By planning to offer open-weight versions, Tencent is signaling a strategy to gain wider developer adoption and integration into local or custom AI applications.
DISCOVERED
7h ago
2026-11-18
PUBLISHED
12h ago
2026-11-18
RELEVANCE