Long-Horizon Terminal-Bench 2.0 (strict pass, reward = 1.0)
Code · 2026-10-09
The share of the 77 Long-Horizon Terminal-Bench 2.0 tasks on which an agent earns the full reward of 1.0, in percent, from the same single 3-hour Terminus 2 attempt per task as the mean-reward headline.
Benchmark source: GitHub
Top models (higher is better)
| Model | Score | Source |
|---|---|---|
| GPT-6 Astra | 42.9 | LHTB 2.0 README leaderboard (GPT-6-astra) · Tencent |
| GPT-5.6 Sol | 37.7 | LHTB 2.0 README leaderboard (GPT-5.6-sol) · Tencent |
| Hy4 Preview | 18.2 | LHTB 2.0 README leaderboard (Hunyuan hy4) · Tencent |
| Kimi K3 | 15.6 | LHTB 2.0 README leaderboard (Kimi K3) · Tencent |
| GLM-5.3 | 11.7 | LHTB 2.0 README leaderboard (GLM-5.3) · Tencent |
| Doubao Seed 2.1 Pro | 10.4 | LHTB 2.0 README leaderboard (Doubao Seed 2.1 Pro) · Tencent |
| DeepSeek V4 Flash Vision Exp | 9.1 | LHTB 2.0 README leaderboard (DeepSeek V4 Flash) · Tencent |
| DeepSeek V4 Pro 0813 | 9.1 | LHTB 2.0 README leaderboard (DeepSeek V4 Pro) · Tencent |
| GLM-5.2 | 7.8 | LHTB 2.0 README leaderboard (GLM-5.2) · Tencent |
| MiniMax M3 | 6.5 | LHTB 2.0 README leaderboard (MiniMax M3) · Tencent |
| Grok 4.6 | 5.2 | LHTB 2.0 README leaderboard (Grok-4.6) · Tencent |
| Kimi K2.6 | 2.6 | LHTB 2.0 README leaderboard (Kimi K2.6) · Tencent |