Tencent Hy4 Flash Matches Claude Haiku 5.5 at 1/18th the Cost Per Task

Photo: Hippopx (public domain). Illustrative image.
What happened. Tencent launched Hy4 Flash on 16 November 2026, a smaller, cheaper sibling of the Hy4 flagship it released less than two weeks earlier. The pitch is not capability — it is cost. Hy4 Flash is a 224-billion-parameter model built on Tencent’s Hunyuan architecture, with 64 billion Engram memory parameters and only 12 billion parameters active per token, a 1-million-token context window, and native multimodality. It is aimed squarely at everyday office work: research, documents, slide decks, spreadsheets and the long, undramatic tasks that fill a workday.
The price is the story. Hy4 Flash lists at $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens — roughly a twelfth of the full Hy4’s list prices. Through 30 November it is free inside WorkBuddy and half price on TokenHub, a launch promotion rather than the permanent price. The model is live now on TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, with OpenRouter next according to Tencent. All of these figures are Tencent’s own, reported by multiple independent outlets; the Tencent blog post itself was not directly accessible for this article, so treat the numbers as vendor-reported until an independent lab reproduces them.
The blind test is the interesting part. Tencent ran a blind test in which 85 WorkBuddy users rated model outputs across 259 everyday tasks in 27 categories. According to Tencent, Hy4 Flash scored 2.79 out of 4 on user satisfaction — effectively tied with Claude Haiku 5.5’s 2.81, ahead of GLM-5.3-Flash’s 2.53, and behind DeepSeek-V4.1-Flash’s 3.09 — while finishing fastest at a median 6.7 minutes per task and costing ¥0.26 per task against Haiku’s ¥4.73, roughly eighteen times cheaper. On WorkIndex, Tencent’s internal benchmark of typical WorkBuddy tasks, Hy4 Flash scored 42.1, edging past Hy4 Preview’s 39.7, at a per-task cost Tencent puts at one-twentieth of the full Hy4 and one-eighth of DeepSeek-V4.1-Flash. WorkIndex is not public, and the comparison is against Tencent’s own models and one DeepSeek model — not an independent evaluation.
The trade-off is real, and Tencent admits it. On Terminal-Bench 4.0, Hy4 Flash scored 19.2 against Hy4’s 44.4 and DeepSeek’s 40.0 — Tencent itself recommends the full Hy4 for complex coding. Its PPT collaborative-editing score fell from 78.9 to 74.0 against the flagship, and automated task scores from 56.2 to 52.2. The model also writes about half as much text per completed task, which is part of why the per-task cost falls so far. Tencent says it plans to release Hy4 Flash as an open-weight model, with no date and no licence yet. For a plain-English explainer of how these per-token prices actually work, see How Does AI API Pricing Work?.
Why it matters. Hy4 Flash is part of a clear second race among the Chinese labs: after the flagship race, the efficiency race. Alibaba’s Qwen4-Max made the same bet a day earlier, claiming a 62% cut in cost per task. The pattern is consistent — the flagship proves the capability, the Flash proves the economics — and for the everyday-work use case Tencent is targeting, the economics may matter more than the benchmark. The honest question is whether the blind-test tie with Haiku holds up outside Tencent’s own users and tasks. Until an independent evaluation lands, Hy4 Flash is a price story with a promising signal, not a proven ranking.
Sources
- Tencent — “Introducing Hy4 flash: Ultimate cost-efficiency for everyday productivity”, Tencent Hy blog (16 November 2026; as reported by AI Frontier Post, Pasquale Pillitteri, dev.to and AshutoshAI — the Tencent blog page itself was not directly accessible)
- AI Frontier Post — “Tencent launches Hy4 Flash: a 224B model at one-twentieth of Hy4’s per-task cost” (16 November 2026): https://aifrontierpost.com/articles/tencent-hy4-flash-launch/
- Pasquale Pillitteri — “Tencent Launches Hy4 Flash, a 1M-Context Model at $0.07 per Million Tokens” (16 November 2026): https://pasqualepillitteri.it/en/news/34117/tencent-hy4-flash-1m-context-model-7-cents
- FreshSynapse — “Qwen4-Max Is Here: Alibaba’s 2.4T-Parameter Flagship Cuts Agent Costs 62%”: https://freshsynapse.com/articles/qwen4-max-alibaba-flagship-ai-teammate-2026-11-16/
- FreshSynapse — “How Does AI API Pricing Work? Token Pricing Explained”: https://freshsynapse.com/articles/how-ai-api-pricing-works-token-pricing-explained/