DeepSeek Deepseek Pro v4

3 appearances · meta-score 58.6/100 · last seen 2026-07-05 · scored under rubric v15

Pooled composite across 3 primary runs: 53.9 ± 5.0 (95% CI)

$67.22Total cost
154,667,711Tokens
82.0%Cache hit
2,366Requests
0.13%Failure rate
1.18Tools/response
2,300,897Tokens per $

Agent role scores

Per-agent-role scores for DeepSeek Deepseek Pro v4
Agent role	Score / 10
Sprint-Plan Agent	7.58
Scout Agent	7.43
Sprint-Review Agent	6.78
Spec Agent	6.60
Test Agent	6.22
Develop Agent	6.14
Code-Review Agent	5.56
Pseudocode Agent	5.25
Platform Agent	5.19
Build Agent	4.89
Code Agent	4.82
Orchestrator	4.68

Run history (primary model)

Runs where DeepSeek Deepseek Pro v4 was the primary model, oldest first
Run	SoW	Date	Composite
Deepseek Pro v4	circles	Jun 2026	53.9
Deepseek Pro v4	circles	Jul 2026	51.9
Deepseek Pro v4	circles	Jul 2026	55.9

Every run on Favur Evals is scored by the same deterministic engine, on the same Statements of Work. The benchmark is self-funded — no vendor sponsorship, credits, or grants.