The Hy3 official version is honed for real-world business scenarios, using a Mixture-of-Experts (MoE) architecture with 295B total parameters and 21B activated parameters. It natively supports a 256K context window and offers multiple thinking modes: no_think (ultra-fast response), think_low (quick thinking), and think_high (deep reasoning), balancing ultra-fast responses, complex reasoning, and invocation cost. Compared with the Preview version, Hy3—based on real business feedback from Tencent Yuanbao, WorkBuddy, ima, Marvis, and others—focuses on improving the Coding Agent, long-form understanding, multi-turn context continuity, search QA, and complex task execution, performing more stably in reducing hallucinations, improving task completion, and engineering usability. It is better suited to practical scenarios such as frontend tasks, cross-file code development, long-document analysis, office automation, and multi-step Agent workflows.
hy3 vs hy3-preview
Output tokens cost $0.62 per million on hy3 and $0.57 per million on hy3-preview. Input tokens cost $0.16 per million on hy3 and $0.17 per million on hy3-preview. Cached input tokens are billed at $0.04 per million on hy3 and $0.05 per million on hy3-preview. Context lengths are 256,000 tokens on hy3 and 256,000 tokens on hy3-preview. Time to first token (TTFT) measured on AIHubMix is 3.3s on hy3 and 2.5s on hy3-preview. Measured output throughput is 33.0 tok/s on hy3 and 49.3 tok/s on hy3-preview. On the LMArena coding leaderboard hy3 scores 1503 and hy3-preview scores 1503.
Hunyuan Hy3 preview is designed for agent workloads, adopting a MoE architecture with 295B capacity and 21B activated parameters. It provides three modes within the same model—no_think (ultra-fast response), think_low (fast thinking), and think_high (deep reasoning)—to accommodate different latency and depth requirements from high-frequency interactions to complex engineering tasks. On code benchmarks such as SWE-bench Verified it approaches the current state of the art, and its 256K context supports cross-file code refactoring and long-document analysis. It is suitable for developers who require reliable task completion while being sensitive to inference costs.
Pricing & Specifications
Prices are per million tokens. Time to First Token and throughput are rolling averages measured on AIHubMix.
Promotional prices show the discounted rate; see each model page for promotion windows.
Activity Past 30 Days
Daily traffic served through AIHubMix — how demand for each model is trending.
Tokens / day
Requests / day
Performance Past 3 Days
Measured on real AIHubMix traffic, hourly buckets. Gaps mean no traffic in that hour.
Throughput (tok/s)
TTFT (s)
Uptime (%)
LMArena Benchmarks
LMArena ratings by capability (Bradley-Terry, commonly called Elo). Higher is better.
Source: LMArena (arena.ai) leaderboard, imported by AIHubMix. Models without published ratings are omitted per chart.
Cost calculator
Estimate your monthly bill for the same workload on each model.
Monthly = daily × 30. Discounted rates applied where a promotion is active.
FAQ
Which is cheaper: hy3, hy3-preview?
hy3-preview: $0.57/M output tokens; hy3: $0.62/M. Use the cost calculator above to estimate your own workload.
How do their coding arena scores compare?
hy3: 1503; hy3-preview: 1503 (LMArena coding leaderboard).
Which responds faster?
hy3-preview: 2.5s time to first token measured on AIHubMix; see the live performance charts above for how each model behaves across the day.
How large is each context window?
hy3 accepts 256,000 and hy3-preview accepts 256,000 input tokens. Maximum output per request is 128,000 tokens on hy3 and 128,000 tokens on hy3-preview.
Which one generates tokens faster?
hy3-preview at 49.3 tok/s and hy3 at 33.0 tok/s, measured as output throughput on AIHubMix — a separate metric from time to first token.
What inputs and capabilities does each model support?
hy3 accepts text input and supports tool calling, function calling, structured outputs, web search, long context and thinking; hy3-preview accepts text input and supports tool calling, function calling, structured outputs, web search, long context and thinking.
Can I call hy3 and hy3-preview with the same API key?
Yes. AIHubMix serves every model on this page behind one OpenAI-compatible endpoint, so switching between them is a one-line change to the model field — no second account, key or SDK.
Popular comparisons
Related model match-ups readers also look at.
