Hy3
Released · tencent.ai
Price
| Per 1M tokens | Input | Cached | Output |
|---|---|---|---|
| List price | ¥1 | ¥0.25 | ¥4 |
Benchmark results
| Benchmark | Score | Source verified |
|---|---|---|
| Humanity's Last ExamNo tools · text-only subset; highest reasoning setting | 37.0% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; text-only subset; highest reasoning setting | 30.0% | Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| re-evaluated at Hy4 preview launch; text-only subset; highest reasoning setting · reported later, in Introducing Hy4 preview | 34.4% | |
| Humanity's Last ExamWith tools · text-only subset; highest reasoning setting | 53.2% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; text-only subset; highest reasoning setting | 35.4% | Corroborated The same number is printed in 1 other lab document: 1 |
| re-evaluated at Hy4 preview launch; text-only subset; highest reasoning setting · reported later, in Introducing Hy4 preview | 51.9% | |
| SWE-bench VerifiedSWE-agent scaffold; highest reasoning setting | 78.0% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; SWE-agent scaffold; highest reasoning setting | 74.4% | Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| SWE-bench ProSWE-agent scaffold; highest reasoning setting · The source doesn't say whether this is the public or the commercial set | 57.9% | Corroborated The same number is printed in 2 other lab documents: 1 2 |
| Hy3 preview, re-evaluated at Hy3 launch; SWE-agent scaffold; highest reasoning setting · The source doesn't say whether this is the public or the commercial set | 46.0% | Corroborated The same number is printed in 2 other lab documents: 1 2 |
| GPQA Diamondhighest reasoning setting | 90.4% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; highest reasoning setting | 87.2% | Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| re-evaluated at Hy4 preview launch; highest reasoning setting · reported later, in Introducing Hy4 preview | 90.9% | |
| BrowseCompinternal harness; self-summary context management; highest reasoning setting | 84.2% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; internal harness; self-summary context management; highest reasoning setting | 67.1% | Corroborated The same number is printed in 3 other lab documents: 1 2 3 |
| AutomationBenchAutomationBench v1.0.6; highest reasoning setting · reported later, in Introducing Hy4 preview | 16.1% | |
| Terminal-Bench 2.1Terminus-2 harness; 4h agent timeout; up to 500 episodes; highest reasoning setting | 71.7% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; Terminus-2 harness; 4h agent timeout; up to 500 episodes; highest reasoning setting | 58.0% | Corroborated The same number is printed in 1 other lab document: 1 |
| re-evaluated at Hy4 preview launch; Claude Code harness; up to 500 turns and a 12h timeout per trial; highest reasoning setting · reported later, in Introducing Hy4 preview | 70.8% | |
| MCP Atlas500-task public set; 100-tool-call budget; Gemini 2.5 Pro judge; highest reasoning setting | 79.1% | Corroborated The same number is printed in 1 other lab document: 1 |
| Hy3 preview, re-evaluated at Hy3 launch; 500-task public set; 100-tool-call budget; Gemini 2.5 Pro judge; highest reasoning setting | 66.1% | Corroborated The same number is printed in 1 other lab document: 1 |
| re-evaluated at Hy4 preview launch; 500-task public set; 100-tool-call budget; Gemini 3.1 Pro Preview judge; highest reasoning setting · reported later, in Introducing Hy4 preview | 75.0% | |
| CyberGymharness not stated; highest reasoning setting · reported later, in Introducing Hy4 preview | 51.8% |
Sources
Reported by Tencent
- Introducing Hy3hy.tencent.ai · 6 Jul 2026
Original image - Introducing Hy4 previewhy.tencent.ai · 28 Aug 2026
Original image
