Tencent AI API

Tencent Hy4 Preview is an open-weight Mixture-of-Experts model from Tencent's Hunyuan team, the successor to Hy3. It has 770B total parameters with 49B active per token, and supports a context window over 1M tokens. It's released under the Apache 2.0 license and supports tool calling. In Tencent's internal blind evaluation across 203 engineering tasks judged by 163 experts, Hy4 Preview scored 2.99 out of 4.00, ahead of GLM-5.3 (2.92) and Kimi K3 (2.94). It scores 85.4 on Terminal-Bench 2.1, surpassing DeepSeek V4 Pro, and 82.9 on SWE-bench Multilingual, up from Hy3's 75.8. On DeepSWE, its score rose from Hy3's 28.0 to 64.3. It's aimed at coding agents, software engineering, office and document generation, and scientific research workflows via API.

Hy-MT2 1.8B is a lightweight multilingual translation model from Tencent Hunyuan, part of the Hy-MT2 family that also includes 7B and 30B-A3B (MoE) variants for heavier workloads. It translates across 33 languages, including Chinese, Cantonese, English, French, Japanese, Korean, Arabic, Russian, Tibetan, and Uyghur, and is trained to follow translation-specific instructions such as terminology guidance, style adaptation, and structured data translation. Tencent reports that the 1.8B model outperforms mainstream commercial translation APIs from providers like Microsoft and Doubao overall, though it hasn't published specific BLEU or COMET scores to back up the comparison. The larger 7B and 30B-A3B models are reported to outperform DeepSeek-V4-Pro and Kimi K2.6 on translation tasks in fast-thinking mode. With an 8,192-token context window, it's a fit for developers who need fast, low-cost, dedicated translation via API rather than a general-purpose chat model.

Hy-MT2 30B-A3B is a multilingual translation model from Tencent's Hunyuan team, built as a Mixture-of-Experts network with 30B total parameters and roughly 3B active per token. It's part of the Hy-MT2 family, released alongside 1.8B and 7B variants, and is designed for fast, instruction-following translation across 33 languages, including style guides, terminology lists, and structured data formats. Tencent reports it outperforming open-source models including DeepSeek-V4-Pro and Kimi K2.6 on translation quality evaluations, and approaching closed-source Gemini 3.1 Pro. It ships under the Apache 2.0 license. It's a fit for developers building translation features who need multilingual quality without the cost of a general-purpose flagship model, particularly when terminology, tone, or output format need to be controlled via instructions.

Hy-MT2 7B is a multilingual translation model from Tencent's Hunyuan team, part of the Hy-MT2 family that also includes the 1.8B and 30B-A3B (MoE) variants. Tencent positions it as the balanced-performance size, between the lightweight 1.8B model and the professional-grade 30B-A3B model. It translates across 33 languages and follows translation-specific instructions for terminology, style, and output format. On the FLORES-200 XX↔XX benchmark, Hy-MT2 7B scores 86.89, reaching 97.9% of Gemini 3.1 Pro's performance. That compares to 87.47 (98.6%) for the 30B-A3B model and 79.77 (89.9%) for the 1.8B model in the same family. Tencent also reports it outperforming DeepSeek-V4-Pro and Kimi K2.6 on translation tasks in fast-thinking mode. With an 8,192-token context window, it's a fit for developers who need higher translation quality than the 1.8B model without the cost of the 30B-A3B variant, particularly for terminology- or format-sensitive translation via API.

Tencent Hy3 is the official release of Tencent Hunyuan's Hy3 series, a hybrid fast-and-slow-thinking Mixture-of-Experts model with 295B total parameters and 21B active per query, following the earlier hy3-preview. It supports a 262K-token context window and up to 131K output tokens. Compared to the preview, Tencent reports agent and coding capability gains of 20-30%, a hallucination rate cut from 12.5% to 5.4%, and commonsense error rates nearly halved. It scores 78 on SWE-bench Verified, 57.9 on SWE-bench Pro, and 90.4 on GPQA Diamond, and Tencent says it matches flagship models with two to five times its parameter count. Hy3 ships under Apache 2.0 and already powers Tencent products like WorkBuddy/CodeBuddy and Yuanbao, with a reported 90% task resolution rate on Tencent's internal WorkBuddy platform. It's a good fit for developers building coding agents, tool-using workflows, and long-context reasoning pipelines who want strong performance at a lower active-parameter cost.

Tencent Hy3 is a 295B-parameter Mixture-of-Experts reasoning model developed by Tencent's Hunyuan team, with only 21B parameters active per query. It supports a 256K-token context window and configurable reasoning levels (disabled, low, high), letting you trade off latency and depth per request. Hy3 is particularly strong on coding and agentic tasks. It scores 74.4% on SWE-bench Verified for real-world bug fixing and 67.1% on BrowseComp for complex web research. Its MoE architecture delivers competitive performance against much larger models — matching Kimi-K2.5 (1T+ parameters) on agent benchmarks at a fraction of the compute cost. Best suited for developers building agentic workflows, code generation pipelines, and multi-step reasoning applications where cost-efficiency matters.

Hunyuan Image 3.0 is a text-to-image generation model from Tencent's Hunyuan team, built as an 80B-parameter Mixture-of-Experts model with 13B parameters active per token. Unlike diffusion-based generators, it uses a unified autoregressive framework modeling text and image tokens together, and Tencent calls it the largest open-source MoE image generation model released to date. The model is tuned for prompt understanding, using world-knowledge reasoning to interpret sparse prompts and fill in contextually appropriate detail, aiming for photorealistic, semantically accurate output. In Tencent's human evaluation of 1,000 prompts by over 100 professional evaluators, Hunyuan Image 3.0 scored a 14.10% relative win rate over its predecessor, Hunyuan Image 2.1, and beat Seedream 4.0 (1.17%), Nano Banana (2.64%), and GPT-Image (5.00%) by the same measure. It's part of the same Hunyuan family as Tencent's A13B and Hy3 chat models, and suits developers who need strong prompt adherence and photorealism from a text-to-image API.

Hunyuan A13B Instruct is an open-source large language model from Tencent built on a fine-grained Mixture-of-Experts (MoE) architecture, with 80B total parameters and 13B active during inference. It natively supports a 256K-token context window. It performs competitively with OpenAI o1 and DeepSeek R1 across math, science, and reasoning benchmarks, scoring 87.3 on AIME 2024, 89.1 on BBH, and 84.7 on ZebraLogic. Hunyuan A13B particularly excels at agentic tasks and tool use, leading on benchmarks like BFCL-v3 (78.3) and ComplexFuncBench (61.2). It's a strong choice for developers building agent workflows, long-context applications, or cost-sensitive reasoning pipelines.