← All models

Hy-MT2-1.8B

Published Publisher Tencent

Positioning and release

Hy-MT2-1.8B is Tencent’s lightweight, 1.8-billion-parameter, fast-thinking multilingual translation model for complex real-world scenarios. Tencent announced the open-source release on May 21, 2026. The model card lists the weights under the Apache 2.0 license. (github.com)

Capabilities and uses

  • Supports translation among 33 languages and multilingual translation instructions. Tencent also describes support for five Chinese minority-language or dialect translation pairs. (github.com)
  • Intended for general, business and domain-specific translation, including instructions involving terminology, context, style and structured data. (github.com)
  • Available for local deployment. Tencent reports that the separate AngelSlim 1.25-bit quantized variant requires 440 MB of storage and delivers a 1.5× inference speed improvement; these figures apply to that variant, not necessarily the original weights. (github.com)

Limits

  • The official repository specifies a maximum context of 8,192 tokens and recommends a maximum output of 4,096 tokens. Tencent Cloud’s hosted hy-mt2-lite service lists an 8K context, 4K maximum input and 4K maximum output; hosted limits should not be assumed to apply to every local deployment. (github.com)
  • Translation and instruction compliance are not guaranteed. Review important translations, terminology and preserved document structure before use. Performance and resource requirements depend on the deployment and quantization.

Official API pricing

Tencent Cloud calls its hosted 1.8B translation offering Hy-MT2-Lite (hy-mt2-lite). The following are hosted-service rates, not charges for downloading the open weights. The domestic and international offerings are separate; the amounts are not currency conversions. (cloud.tencent.com)

  • China domestic site, Guangzhou, postpaid: CNY 0.3 per million input tokens and CNY 1.2 per million output tokens. The listed row specifies no context-length tier or peak/off-peak rate. Official domestic pricing, checked November 16, 2026. (cloud.tencent.com)
  • International site, listed Singapore and Guangzhou regions: USD 0.044 per million input tokens and USD 0.177 per million output tokens. The listed rows specify no context-length tier or peak/off-peak rate. Official international pricing, checked November 16, 2026. (tencentcloud.com)

Neither listing specifies a separate cache rate for these rows. Confirm region availability and the final bill in the relevant service console.

Official sources