TencentHy3
As of 29 Oct 2026, Hy3 by Tencent scores 55.9 on the BenchLeader Index (581 ranked configurations), with its best results in maths (67) and weakest in agents & tools (47). It costs $0.243 per 1M tokens blended ($0.140 in / $0.550 out), and streams about 82 tokens/s (Artificial Analysis). Released 6 Jul 2026; 9 independent results so far.
Key facts (as of 29 Oct 2026)
- Hy3 costs $0.243 per 1M tokens blended ($0.140 in / $0.550 out) (list price, 29 Oct 2026).
- Hy3 streams about 82 tokens/s (Artificial Analysis, 29 Oct 2026).
- Hy3 was released on 6 Jul 2026 (Tencent).
In detail. Hy3 is a Tencent open-weights model, released 6 Jul 2026. Its best configuration ranks #185 of 581 on the BenchLeader Index at 55.9 ±6.6, in the upper half. The ± is the point: 320 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in maths (67) and lowest in agents & tools (47). At $0.243 per million tokens blended it is among the cheapest fifth of ranked models. Output speed of 82 tokens per second puts it slower than most, with a first answer in 27.1 s. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 28 Oct 2026.
- Blended price
- $0.243/M
- $0.140 in · $0.550 out
- Output speed
- 82 tok/s
- measured by Artificial Analysis
- First answer
- 27 s
- first token 2.67 s
- Context
- 262k
- Full answer
- 33 s
- median, reasoning included
- Cost per run
- –
- Released
- 6 Jul 2026
- Overall index56
- Reasoning48
- Coding57
- Agents & tools47
- Maths67
- Human preference64
- Long context64
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | – | – | 82 tok/s | 27 s | $0.0002 | Agents & tools 62 · Instruction following 50 · Long context 45 |
| thinking | – | – | 82 tok/s | 27 s | $0.0002 | Agents & tools 63 · Instruction following 63 · Long context 57 |
| not statedbest | 55.9 | #185 | 82 tok/s | 27 s | $0.0002 | Agents & tools 47 · Coding 57 · Human preference 64 · Long context 64 · Maths 67 · Reasoning 48 |
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Hard Promptseffort not stated | – | – | 1476#73 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 73.2%#275 | 86.7%#109 | 89.7%#71 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 7.0%#385 | 27.8%#181 | 33.5%#142 | Artificial Analysis | |
| HLE-Diamond (AA)effort not stated | – | – | 11.6%#90 | Artificial Analysis | |
| CritPtnot in index | 0.3%#283 | 4.6%#172 | 4.9%#167 | Artificial Analysis |
Coding
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Coding | 1503#73 | Arena (LMArena) | |
| Arena WebDev | 1505#63 | Arena (LMArena) | |
| SciCode (AA)not in index | 48.6%#130 | Artificial Analysis | |
| Terminal-Bench 4.0 (AA)not in index | 0.5%#149 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 64.4%#86 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Agenteffort not stated | – | – | -7.3#40 | Arena (LMArena) | |
| Terminal-Bench Hard | 31.8%#102 | 34.1%#89 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 67.5%#165 | 92.7%#51 | – | Artificial Analysis | |
| GDPval-AA v2.1not in indexeffort not stated | – | – | 1071#90 | Artificial Analysis | |
| τ³-Banking (AA)effort not stated | – | – | 22.9%#90 | Artificial Analysis | |
| AutomationBenchnot in indexeffort not stated | – | – | 18.4%#149 | Artificial Analysis | |
| GDP.pdf (AA)not in indexeffort not stated | – | – | 11.2%#130 | Artificial Analysis | |
| AA-Briefcase v1.1not in indexeffort not stated | – | – | 947#103 | Artificial Analysis |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Maths | 1485#36 | Arena (LMArena) |
Instruction following
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 48.0%#188 | 63.1%#116 | – | Artificial Analysis | |
| Arena Instruction Followingnot in indexeffort not stated | – | – | 1450#67 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1455#74 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1426#77 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1451#87 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1467#66 | Arena (LMArena) |
Long context
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR v1.1 | 42.0%#362 | 64.7%#264 | 79.0%#112 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA Intelligence Index v4.3.2not in index | 17.0#273 | 22.7#206 | 25.3#169 | Artificial Analysis | |
| AA Progress Index v1not in index | 233.6#304 | 267.3#206 | 285.3#151 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| NovitaAI | 81 tok/s | 3.17 s | $0.140 | $0.580 | 262k | – |
| AtlasCloud | 80 tok/s | 1.58 s | $0.200 | $0.800 | 262k | fp8 |
| Tencent Cloud | 66 tok/s | 2.67 s | $0.083 | $0.330 | 262k | fp8 |
| GMICloud | 57 tok/s | 2.46 s | $0.140 | $0.580 | 262k | bf16 |
| Phala | 37 tok/s | 1.66 s | $0.150 | $0.640 | 262k | – |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.033 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0002 | $0.0002 | 30.8 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0020 | $0.0010 | 34.4 s |
| Code edit | 6,000 / 1,500 | $0.0017 | $0.0012 | 45.4 s |
| Agentic coding session | 60,000 / 4,000 | $0.011 | $0.0058 | 1.3 min |
| Structured extraction | 2,000 / 200 | $0.0004 | $0.0002 | 29.5 s |
More from Tencent
See also
Questions people ask
- What are Hy3's benchmark scores?
- As of 29 Oct 2026, Hy3 scores 55.9 ±6.6 on the BenchLeader Index, strongest in maths (67) and weakest in agents & tools (47). That rests on 9 independent benchmark results, and every score on this page links to its publisher.
- How does Hy3 rank?
- Hy3 is #185 of 581 ranked configurations on 29 Oct 2026. 320 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 69.1.
- How much does Hy3 cost per 1M tokens?
- As of 29 Oct 2026, Hy3 costs $0.140 per 1M input tokens and $0.550 per 1M output tokens, $0.243 blended (three input tokens per output token).
- When was Hy3 released?
- Tencent released Hy3 on 6 Jul 2026. BenchLeader has results for it at 3 reasoning-effort settings.
Data as of 29 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Hy3: benchmarks, pricing, speed and rank”, data as of 29 Oct 2026, https://www.benchleader.com/models/hy3.