Tencent: Hy3
tencent/hy3
49th smartest of 151Cheaper than 76% of paidReasoningToolsJSON
Intelligence
33.6
49th of 151
Design Elo
—
Speed
36
227th fastest
Latency
452ms
first token
Input price
$0.140
73rd cheapest
Context
262K
131K max out
How it compares
Smarter than68%
of all ranked models
Faster than28%
of all ranked models
Cheaper than76%
of all ranked models
Overview
Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort:...
Benchmarks
independent · Artificial Analysis & Design ArenaArtificial Analysis85th percentile
Intelligence Index
33.6
GPQA Diamond
87%
Humanity's Last Exam
26%
SciCode
41%
Tau²-Bench (agentic)
93%
Providers & pricing (1)
| Provider | In $/M | Out $/M | Context | Uptime |
|---|---|---|---|---|
| AtlasCloudfp8 | $0.140 | $0.580 | 203K | 100% |
Specifications
Context window262K
Max output131K
Knowledge cutoff—
Input modalitiestext
Output modalitiestext
Prompt caching—
Cache read price$0.035/M
ModeratedNo
Open weightstencent/Hy3 ↗
Hy3 FAQ
How much does Hy3 cost?
Hy3 costs $0.140 per million input tokens and $0.580 per million output tokens via OpenRouter, making it 73rd cheapest of 302 paid models.
How smart is Hy3?
Hy3 scores 33.6 on the Artificial Analysis Intelligence Index, ranking 49th of 151 benchmarked models, with a GPQA Diamond score of 87%.
How fast is Hy3?
Hy3 generates around 36 tokens per second with 452ms time-to-first-token (p50), the 227th fastest tracked model.
What is Hy3's context window?
Hy3 supports a 262K-token context window and can output up to 131K tokens. It accepts text input.