Tencent's pricing move: Hy4 Flash undercuts Meta, Anthropic and OpenAI on listed rates, with all the evidence so far internal.

Tencent released Hy4 Flash on Nov. 16, an AI model built for everyday office work and priced at $0.070 for a million input tokens and $0.281 for a million output tokens

1

2

. The price is the pitch: the Chinese technology giant says the model undercuts its main US rivals on their own listed rates

1

.

Per Tencent's comparison, Meta's Muse Ember, Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna list base rates of $0.10 per million tokens in and $0.50 out

1

. They charge more once a job runs past their long-context thresholds

1

. The model is available through Tencent Cloud's TokenHub, the WorkBuddy productivity app and the Yuanbao assistant, with free WorkBuddy access and a 50% API discount through Nov. 30

1

2

.

Why the model can be cheap

Hy4 Flash has 224 billion total parameters, with 12 billion activated per request

1

2

. Per Tencent, that sparse design is what makes the low price technically possible

1

. The company also reports speeds of more than 150 tokens per second

1

2

.

The context window tops out at a million tokens, meaning the amount of text the model can work with at once

1

2

. Long documents are where Tencent says its rivals' pricing climbs, so the million-token window at this rate is aimed at that

1

.

The evidence is all internal

In an internal blind test, 85 WorkBuddy users completed 259 tasks across 27 categories

1

2

. Tencent says its own blind test put Hy4 Flash at 2.79 on user satisfaction against 2.81 for Claude Haiku 5.5, at roughly one-eighteenth the per-task cost

1

2

. By its own internal WorkIndex benchmark, Tencent puts the model's per-task cost at a twentieth of flagship Hy4's and a thirteenth of Hy4 Preview's

1

2

.

The benchmarks and blind evaluations behind those quality and cost figures were all run by Tencent itself

1

2

.

A budget tier, and a fast cadence

Tencent itself positions Flash as the budget tier. It recommends the model for document processing, slide creation and research, and sends complex software engineering to the flagship Hy4

1

2

. Hy4 Flash is the family's third release since August, after Hy4 Preview on Aug. 28 and the flagship on Nov. 4

1

2

. The company says the family now balances intelligence and efficiency better for real productivity tasks, at what it calls the lowest cost of any Hy model to date

2

.

Tencent plans to open-source the model's weights, with availability through OpenRouter coming soon

1

2

.

© 2026 TheOutpost.AI All rights reserved