Compare / head-to-head
Hy3vsHy4 preview
Hy4 preview leads 9 of 9 shared benchmarks. Hy3 is 5.1x cheaper per token. Hy4 preview has the larger context window (1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 9
Hy4 preview leads
Cheaper per token
Hy3
5.1x cheaper, input + output
Larger context
Hy4 preview
1M tokens
Providers
Tencent Hunyuan
same provider
Hy3
Tencent Hunyuan · released 2026-07-06
text
Hy4 preview
Tencent Hunyuan · released 2026-08-28
text
Quality
Benchmark matrix
9 shared · 35 only Hy3 · 9 only Hy4 preview| Benchmark | Hy3 | Hy4 preview | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 9 | ||||
| GPQA Diamond | 90.4% ↗ | 92.3% ↗ | -1.9 pt | Hy4 preview |
| DeepSWE | 28% ↗ | 64.3% ↗ | -36.3 pt | Hy4 preview |
| HLE (with tools, text-only) | 53.2% ↗ | 55.4% ↗ | -2.2 pt | Hy4 preview |
| LMArena Agent (LMArena) | -0.0538 ↗ | 0.0395 ↗ | -0.1 | Hy4 preview |
| LMArena WebDev (LMArena) | 1507.4 ↗ | 1633.2 ↗ | -125.8 | Hy4 preview |
| SWE-bench Multilingual | 75.8% ↗ | 82.9% ↗ | -7.1 pt | Hy4 preview |
| SWE-bench Pro | 57.9% ↗ | 65.7% ↗ | -7.8 pt | Hy4 preview |
| Terminal-Bench 2.1 | 71.7% ↗ | 85.4% ↗ | -13.7 pt | Hy4 preview |
| WideSearch | 76.4% ↗ | 83.9% ↗ | -7.5 pt | Hy4 preview |
| Only Hy3 reports · 35 | ||||
| SWE-bench Verified | 78% ↗ | not reported | — | — |
| Humanity's Last Exam (no tools) | 37% ↗ | not reported | — | — |
| LMArena Elo | 1440.6 ↗ | not reported | — | — |
| AA-LCR | 73.4% ↗ | not reported | — | — |
| Apex-Agent | 25.6% ↗ | not reported | — | — |
| ArxivMath | 52.2% ↗ | not reported | — | — |
| BrowseComp | 84.2% ↗ | not reported | — | — |
| CL-bench | 23.8% ↗ | not reported | — | — |
| CL-bench Life | 17% ↗ | not reported | — | — |
| ClawEval | 68.5% ↗ | not reported | — | — |
| CMT-Benchmark | 37.8% ↗ | not reported | — | — |
| DeepSearchQA | 91% ↗ | not reported | — | — |
| e-bench | 50.2% ↗ | not reported | — | — |
| FrontierScience-Olympiad | 74.8% ↗ | not reported | — | — |
| FrontierScience-Research | 21.3% ↗ | not reported | — | — |
| HorizonMath | 7.1% ↗ | not reported | — | — |
| Hy-Backend 2.0 | 25% ↗ | not reported | — | — |
| Hy-CompanyBench | 41.7% ↗ | not reported | — | — |
| Hy-Euler Pro | 24.2% ↗ | not reported | — | — |
| Hy-FinModelBench | 69% ↗ | not reported | — | — |
| Hy-Math | 60.9% ↗ | not reported | — | — |
| Hy-SkillsWorld | 45.8% ↗ | not reported | — | — |
| Hy-SWE Max | 49% ↗ | not reported | — | — |
| IMOAnswerBench | 90% ↗ | not reported | — | — |
| MathArena Apex | 38.7% ↗ | not reported | — | — |
| MCP Atlas | 79.1% ↗ | not reported | — | — |
| NL2Repo | 45.6% ↗ | not reported | — | — |
| PHYBench | 77.4% ↗ | not reported | — | — |
| ProdBench | 23% ↗ | not reported | — | — |
| SkillsBench | 55.3% ↗ | not reported | — | — |
| SuperChem | 54.9% ↗ | not reported | — | — |
| Toolathlon | 48.5% ↗ | not reported | — | — |
| Toolathlon-Verified (HKUST) | 56.8% ↗ | not reported | — | — |
| USAMO 2026 | 72% ↗ | not reported | — | — |
| WildClawBench | 53.6% ↗ | not reported | — | — |
| Only Hy4 preview reports · 9 | ||||
| Agents' Last Exam (ALE-CLI) | not reported | 22.8% ↗ | — | — |
| CyberGym | not reported | 78.4% ↗ | — | — |
| HLE (no tools, text-only) | not reported | 43.4% ↗ | — | — |
| MathArena Apex 2025 | not reported | 74.2% ↗ | — | — |
| MCP-Atlas (public) | not reported | 83.7% ↗ | — | — |
| NL2Repo-Bench | not reported | 58.9% ↗ | — | — |
| OfficeQA Pro | not reported | 66.2% ↗ | — | — |
| Terminal-Bench 4.0 (Vals AI) | not reported | 8.1% ↗ | — | — |
| Toolathlon-Verified | not reported | 74.1% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Hy3 minus Hy4 preview in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Hy3 | Hy4 preview | Edge |
|---|---|---|---|
| Context window | 256K tokens | 1M tokens | Hy4 preview |
| Max output | — | 64K tokens | — |
| Input price / 1M | $0.132 ↗ | $0.834 ↗ | Hy3 |
| Output price / 1M | $0.528 ↗ | $2.5 ↗ | Hy3 |
| Cached input / 1M | $0.033 ↗ | $0.042 ↗ | Hy3 |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $0.66 | $3.34 | Hy3 |
| Modalities | text | text | Tie |
| Released | 2026-07-06 | 2026-08-28 | — |
| Cited benchmark scores | 45 | 18 | — |
Reliability
Tencent Hunyuan status
Live from /statusMore matchups
Hy3 vs …
Models sharing the most benchmarksMore matchups
Hy4 preview vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Hy3 and Hy4 preview on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.