Tencent Hy3
Tencent · Released Jul 2026 · hy3
$0.132
Input price / 1M tokens
#2 of 61, cheapest first
$0.528
Output price / 1M tokens
#3 of 61, cheapest first
262K
Context window
#54 of 71
87
Output tokens / second
#37 of 61
57.9%
SWE-bench Pro
#12 of 16
71.7%
Terminal-Bench 2.1
#16 of 17
53.2%
Humanity’s Last Exam (with tools)
#15 of 16
84.2%
BrowseComp
#5 of 7
Summary
Tencent Hy3 is an open-weight model from Tencent, released on Jul 2026.
At $0.132 input and $0.528 output per million tokens, it is #2 of 61 on input price, cheapest first.
Its 262K-token context window ranks #54 of 71.
Measured output speed is 87 tokens per second, #37 of 61.
Tencent Hy team's MoE model, 295B total and 21B active parameters (plus a 3.8B MTP layer), the full release after Hy3 Preview (late April 2026). Apache-2.0 with no extra conditions, so commercial use is unrestricted.
Benchmarks · 5 reported
| Benchmark | Score | Bar | Rank | Source |
|---|---|---|---|---|
| SWE-bench Proread from benchmark chart image | 57.9% | #12 of 16 | tencent/Hy3 model cardSelf-reported | |
| SWE-bench Multilingualread from benchmark chart image | 75.8% | – | tencent/Hy3 model cardSelf-reported | |
| Terminal-Bench 2.1read from benchmark chart image | 71.7% | #16 of 17 | tencent/Hy3 model cardSelf-reported | |
| Humanity’s Last Exam (with tools)with tools, text-only; read from chart image | 53.2% | #15 of 16 | tencent/Hy3 model cardSelf-reported | |
| BrowseCompread from benchmark chart image | 84.2% | #5 of 7 | tencent/Hy3 model cardSelf-reported |
Self-reported means the lab ran the test itself. A rank appears only where several labs report the same version of a benchmark.
Similar models
Models with the most benchmarks in common with Tencent Hy3, and the closest scores.
SWE-bench Pro
#12 of 16Harder multi-file engineering tasks
- Tencent Hy357.9
- GPT-5.558.6
- GLM-5.262.1
- Qwen3.8-Max67.7
- GPT-5.6 Sol64.6
- Kimi K2.658.6
Terminal-Bench 2.1
#16 of 17Terminal tasks, earlier set
- Tencent Hy371.7
- GLM-5.281
- Qwen3.8-Max86.6
- GPT-5.6 Sol88.8
Humanity’s Last Exam (with tools)
#15 of 16Expert questions, with search and code
- Tencent Hy353.2
- GPT-5.552.2
- GLM-5.254.7
- Qwen3.8-Max56.2
- Kimi K2.654
Compare Tencent Hy3 side by side
Pre-filled with the two closest models. Swap any of them.
Sources
- price (TokenHub Singapore): tencentcloud.com/document/product/1300/78937, checked 2 Oct 2026
- API model ID, context, max input, max output: tencentcloud.com/document/product/1300/78934, checked 2 Oct 2026
- parameters, context 256K, license, benchmarks: huggingface.co/tencent/Hy3, checked 2 Oct 2026
- context 262,144 (max_position_embeddings): huggingface.co/tencent/Hy3/blob/main/config.json, checked 2 Oct 2026
- OpenRouter price history (8 Jul 2026 archive): web.archive.org/web/20260708153506/https://openrouter.ai/tencent/hy3, checked 2 Oct 2026
- output speed: artificialanalysis.ai/models/hy3, checked 2 Oct 2026
Building on Tencent Hy3?
Get the architecture right before the bill arrives.
We size caching, routing and fallbacks for your workload on a free call, and tell you where a cheaper model is good enough.