KiloBench - Terminal Bench 2.0: leaderboard
Kilo reruns all 89 Terminal-Bench 2.0 tasks in its own production agent harness and averages task completion over five attempts per model; score is completion in percent, higher is better. Same tasks as other Terminal-Bench 2.0 boards under a different harness; Kilo's auto router is excluded.
Metric: Task completion in Kilo's agent harness, mean of 5 attempts (%). Source: kilo.ai. Saturation forecast: Around February 2027. 37 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra | 79.33 | #6 |
| 2 | GPT-5.6 Sol | 76.18 | #22 |
| 3 | Claude Fable 5.1 | 76.18 | #10 |
| 4 | Gemini 3.8 Flash | 75.28 | #30 |
| 5 | DeepSeek V4.1 Flash | 75.28 | #37 |
| 6 | GPT-5.5 | 74.16 | #33 |
| 7 | Grok 4.6 | 73.03 | #29 |
| 8 | Kimi K3 | 72.81 | #35 |
| 9 | Claude Opus 5 | 71.46 | #16 |
| 10 | Claude Fable 5 | 71.01 | #20 |
| 11 | Grok 4.5 | 70.79 | #54 |
| 12 | Claude Opus 4.7 | 70.11 | #51 |
| 13 | Claude Opus 4.8 | 67.64 | #39 |
| 14 | Gemini 3.5 Flash | 64.72 | #66 |
| 15 | Kimi K2.7 Code | 60.67 | #95 |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=kilobench-terminal-bench-2-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.