LLM Stats (Terminal-Bench 4.0): leaderboard
LLM Stats mirror of Terminal-Bench 4.0: 66 containerized terminal tasks weighted toward science-adjacent and frontier engineering work. Relative to earlier releases it lengthens timeouts and raises RAM and CPU on selected tasks, so a failure is more likely the agent’s than the harness’s.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 | 70.6 | #11 |
| 2 | Claude Opus 5.5 | 66.4 | #3 |
| 3 | Claude Mythos 5.1 | 60.9 | #1 |
| 4 | GPT-6 Astra | 57.7 | #6 |
| 5 | Gemini 4 Argon | 57.4 | #8 |
| 6 | Muse Wonder Full | 57.1 | #10 |
| 7 | Claude Fable 5.1 | 55.8 | #9 |
| 8 | Muse Wonder | 52 | #15 |
| 9 | Claude Opus 5 | 51.8 | #12 |
| 10 | Claude Fable 5 | 44.5 | #16 |
| 11 | DeepSeek V4.1 Flash (1017) | 42.3 | #20 |
| 12 | GLM-5.3 | 41.8 | #39 |
| 13 | Claude Haiku 5.5 | 39.2 | #40 |
| 14 | MiniMax-M3.1-Flash | 38.1 | #33 |
| 15 | Grok 4.7 | 38 | #25 |
No result here: #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-terminal-bench-4-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.