Today's price · per 1M tokens
Input
$0.132
per 1M tokens
Output
$0.528
per 1M tokens
Blended
$0.231
blended $/1M — 3:1 weighted input:output
Cached input
$0.033
25% of input — prompt caching
How this price is scoped: Tencent's own price, quoted by the FP8 endpoint Tencent operates for this model ($0.132 in / $0.528 out / $0.033 cache-hit per 1M tokens, read 2026-10-19 ~00:20Z and again 14:33Z). Tencent's RMB rate card carries the matching flat GA "Hy3" row at ¥1 / ¥4 / ¥0.25, re-read 2026-10-19 and unchanged; all three USD prices imply 7.576 CNY/USD to three figures. That is Tencent's own commercial rate for this model, not a market rate, and it differs from the 7.197 implied by Hy4 preview's endpoint prices. We publish what Tencent charges in USD, not a spot re-conversion of its domestic card. Until 2026-10-19 this row instead published the card at a spot rate (6.8031 CNY/USD, 2026-07-07): $0.147 / $0.588 / $0.037.
Price receipt
We read Tencent's own pricing page on and found Hy3 listed with a price on the same row — that read is where the number above comes from. The page prices it in CNY; the USD above is converted from that list at a dated FX rate — see the price note for the figures. We keep the page text we read; its content hash is c0e378884d.
Overview
Tencent Hunyuan Hy3 — a 295B-parameter MoE (~21B active) agentic model, generally available on Tencent Cloud TokenHub since its official 2026-07-07 release, which superseded the Apr 2026 Hy3 preview build. Tencent lists the price in RMB (¥1 in / ¥0.25 cached / ¥4 out per 1M tokens); the USD shown is Tencent's own USD price, quoted by the FP8 endpoint Tencent operates for this model. Context window 262,144 (256K) per Tencent Cloud / OpenClaw provider config, not press copy.
Capabilities
struck through = not supportedBenchmark performance
accuracy % · higher is betterSources: Tencent model card (GPQA Diamond, SWE-Bench Verified) · Tencent model card (asterisked) (Humanity's Last Exam)
Every per-benchmark score we hold for this model is reported by its own vendor, so none is counted toward an accuracy average or any ranking. The percentile above is its standing across the independent composites below.
- GDPval-AA v2: 1068.9 ELO (Real-world work tasks, human baseline = 1000)
- Arena (arena.ai): 1454.2 ELO (Human preference)
- Arena WebDev: 1505.8 ELO (Human preference on building web apps)
- Arena Agent: -0.086 (Score from real agent sessions (IPS); higher is better)
- Arena Agent cost per task: $0.08 (Median measured cost of one real Agent Arena task)
- HLE-Diamond: 11.6% (measured 2026-11-22) — A revised 1,000-question set of Humanity's Last Exam, half reasoning, half knowledge
- GDP.pdf: 11.2% (measured 2026-11-22) — Tasks over long professional documents; share of tasks that pass every criterion
- AutomationBench-AA: 18.4% (measured 2026-11-22) — Multi-step work across simulated business apps, without breaking the rules set
- GPQA Diamond: 89.7% (measured 2026-11-22)
- AA-LCR (long-context reasoning): 79% (measured 2026-11-22)
- CritPt (research-level physics): 4.9% (measured 2026-11-22)
- Terminal-Bench 2.1: 64.4% (measured 2026-11-22)
- Terminal-Bench 4.0: 0.5% (measured 2026-11-22)
- SciCode: 48.6% (measured 2026-11-22)
- Measured by AA: 70 tokens/sec output, 4.51s to first token, 43.53s to first answer token
- Ranks #80 of 207 comparably-measured models by percentile score, across 2 independent measurements
- Ranks #12 of 207 comparably-tested models by normalized performance per dollar
70 tokens/sec output 4.51s latency to first token (TTFT)
Throughput: Artificial Analysis, measured 2026-11-14 · Source: Artificial Analysis (artificialanalysis.ai)
Source: Artificial Analysis, LMSYS Chatbot Arena (UC Berkeley), Tencent model card · updated · See full rankings →
Specifications
- Provider
- Tencent
- Context window
- 262K tokens
- Modality
- text
- Parameters
- Proprietary
- Open source
- No — proprietary
- Released
- Status
- Current
- Last updated
- Tags
Availability verified: — listed on Tencent's own page