← Gemini 3.1 Pro · API · High thinking
Evidenceacceptedvendor_reported
Terminal-Bench
Google DeepMind — Gemini 3.1 Pro · Terminal-Bench 2.0
ev_2026_07_gem31pro_terminal_vendor
Result
68.5%
Confidence B
- Tested
- 2026-02-19
- Runs
- 1
Configuration
How it was measured
- Exact model
- gemini-3.1-pro-preview
- Reasoning
- high
- Agent harness
- Terminus-2
- Tool access
- terminal / shell tools
- Prompting
- vendor terminal agent eval
Stretch STR/tools — vendor Terminal-Bench 2.0 (Terminus-2 harness).
Caveats
Known limitations
- Vendor TB 2.0; peers often cite AA TB 2.1 — not fully comparable.
- Trails Sol AA TB 2.1 88% and Opus AA 84.6% on independent boards.
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.90
quality0.55
recency0.85
health0.85
comparability0.70
reliability0.70
Benchmark
Terminal-Bench
Agentic terminal / tool-use suite measuring multi-step shell and environment task completion (Terminus-2 harness commonly cited).
Benchmark source →