Tencent Hy3 preview places in the top three on 2 benchmarks.
Tencent Apr 2026 Open weights Flagship
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LongBench v2 | 65.4% | #4 of 35 | Lab model card | |
| CL-bench | 22.8% | #3 of 24 | Official leaderboard | |
| AdvancedIF | 79.5% | #2 of 6 | Lab model card | |
| LMArena (Elo) | 1,409 | #122 of 335 | Official leaderboard | |
| CL-bench Life | 15.7% | #6 of 16 | Official leaderboard | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | 66.3% | #15 of 38 | Lab model card | |
| Humanity's Last Exam | 30% | #81 of 154 | Lab model card | |
| WideSearch | 70.2% | #13 of 19 | Lab model card | |
| BullshitBench V2 | 16% | #102 of 129 | Official leaderboard | |
| Extended NYT Connections | 17.2% | #89 of 103 | Official leaderboard | |
| BullshitBench | 12.7% | #95 of 107 | Official leaderboard | |
| CompassArena | 963 | #10 of 10 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GPQA Diamond | 87.2% | #58 of 302 | Lab model card | |
| IMO-AnswerBench | 84.3% | #15 of 37 | Lab model card | |
| FrontierScience | 70% | #6 of 8 | Lab model card | |
| Verifying the Verifiers (Project APE CRED benchmark) | 50.5% | #18 of 23 | Official leaderboard | |
| ResearchClawBench | 12.9 | #22 of 23 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 74.4% | #51 of 155 | Lab model card | |
| Code Arena | 1,358 | #99 of 128 | Official leaderboard | |
| Monthly-SWEBench | 38% | #10 of 12 | Official leaderboard | |
| Hy-SWE Max | 30% | #3 of 3 | Lab model card | |
| Hy-Backend | 54.7% | #1 of 1 | Lab model card | |
| Hy-Vibe Bench | 50.5% | #1 of 1 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| PACT | 1,520 | #13 of 26 | Official leaderboard | |
| Terminal-Bench 2.0 | 54.4% | #32 of 63 | Lab model card | |
| SnakeBench | 24 | #96 of 188 | Official leaderboard | |
| BrowseComp | 67.1% | #51 of 81 | Lab model card | |
| Claw-Eval Pass^3 | 55% | #10 of 13 | Lab model card | |
| WildClawBench | 45.3% | #5 of 5 | Lab model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FORTE | 49.7% | #20 of 21 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 72.3 | #89 of 185 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CompassRank | 73.6 | #13 of 14 | Official leaderboard |
The Self-Reported Index fits one score to the 11 capability results Tencent reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 146.0 to 149.7. How we compute it.
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
The benchmarks the lab chose to headline, with the scores it gave.