← DeepSeek V4 Pro · API · Max effort
Evidenceacceptedvendor_reported
Terminal-Bench
DeepSeek V4 materials — Terminal Bench 2.0 (Pro-Max)
ev_2026_07_dsv4pro_terminal_vendor
Result
67.9%
Confidence B
- Tested
- 2026-04-24
- Runs
- 1
Configuration
How it was measured
- Exact model
- deepseek-v4-pro
- Reasoning
- max
- Agent harness
- vendor_terminal
- Tool access
- terminal agent tools
- Prompting
- vendor published terminal agent
STR stretch + existing_codebase skill support — vendor Terminal-Bench 2.0.
Caveats
Known limitations
- Vendor Terminal-Bench 2.0 not AA Terminus-2 TB 2.1.
- Trails frontier closed on terminal (vendor peers GPT-5.4 xHigh ~75).
Weighting
Evidence weight factors
The engine weights each evidence row by these factors (0–1) when fusing scores — higher is more trusted.
relevance0.85
quality0.55
recency0.90
health0.85
comparability0.65
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 →