Tencent says its Hy4 Flash AI model matches Claude Haiku 5.5 at 18x lower cost Tencent launched Hy4 Flash on November 16, 2026, a 224-billion-parameter model with 12 billion parameters active per inference, and said its internal evaluation of 259 tasks run by 85 WorkBuddy users put Hy4 Flash at 2.79 on user satisfaction versus 2.81 for Anthropic's Claude Haiku 5.5 at roughly 18 times lower cost per task. Tencent priced Hy4 Flash API access at RMB 0.5 per million input tokens and RMB 2 per million output tokens, with cached input at RMB 0.02 per million and a 50% discount through November 30, 2026. The model, the third Hy4 release after Hy4 Preview on August 28, 2026 and the flagship Hy4 on November 4, 2026, supports a 1-million-token context window, multimodal input and decoding above 150 tokens per second, and Tencent said it plans to release the model weights and offer Hy4 Flash through OpenRouter. Tencent says its Hy4 Flash AI model matches Claude Haiku 5.5 at 18x lower cost Photo: Tim Witzdam / Pexels Tencent's internal tests put its new budget model within a hair of Anthropic's Claude Haiku 5.5 on user satisfaction at a fraction of the price per task Tencent says its new Hy4 Flash model performs about as well as Anthropic https://cryptobriefing.com/markets/anthropic/ ’s Claude Haiku 5.5 on everyday work. The company also says it costs roughly 18 times less per task to run. What Tencent actually launched Hy4 Flash went live on November 16, 2026, as the lower-cost member of Tencent’s Hunyuan model family. It is built for the unglamorous side of AI: editing documents, doing research and similar daily productivity chores. The model carries 224 billion parameters in total. Only 12 billion of them switch on for any given inference. Tencent uses a technique it calls LatentMixture of Experts to route each request to a small slice of the network, which keeps compute bills down. The model also uses a hybrid attention mechanism and megakernels, a low-level optimization approach. Tencent reports decoding speeds above 150 tokens per second. Hy4 Flash supports a 1-million-token context window and accepts multimodal input. The Claude comparison, and the fine print The headline claim comes from Tencent’s internal evaluation. The company had 85 users of WorkBuddy, its productivity product, run 259 tasks spread across 27 categories. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Hy4 Flash scored 2.79 on user satisfaction. Claude Haiku 5.5 scored 2.81. The cost gap, by Tencent’s math, is far easier to spot: Hy4 Flash came in roughly 18 times cheaper per task. Tencent also benchmarked Hy4 Flash against its own siblings. The company reports a per-task cost of 1/20th that of the flagship Hy4 model and 1/13th that of Hy4 Preview. API pricing is set at RMB 0.5 per million input tokens and RMB 2 per million output tokens. Cached input tokens cost RMB 0.02 per million. Tencent is also running a 50% discount on those rates through November 30, 2026. A fast-moving release schedule Hy4 Flash is the third Hy4 release in a short window. Tencent shipped Hy4 Preview on August 28, 2026, then followed with the flagship Hy4 on November 4, 2026. Flash arrived just 12 days after the flagship. The model is already running inside WorkBuddy and Yuanbao, two of Tencent’s consumer-facing products. It is also set to become available through OpenRouter, a platform developers use to access many AI models through a single interface. Tencent has additionally announced plans to release the model weights, which would let developers download and run Hy4 Flash themselves rather than paying per call through an API. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .