DeepSeek V4 Flash · High

Data published

DeepSeek V4 Flash · High

Reasoning
High reasoning
Made by
DeepSeek

Sent as medium. Fireworks documents that for DeepSeek V4 "'medium' is silently promoted to 'high'" (docs.fireworks.ai chat completions reference, checked 2026-10-02), so this configuration ran at high.

AI Safety Index

48.82 / 100

16th of 17 published models

10 of 10 benchmarks · 100.0% of configured weight

Benchmark profile

Out of 100; higher is safer.

Benchmark scores for DeepSeek V4 Flash · High
BenchmarkScore visualizationScore out of 100
SIM-VAILMental & Emotional Safety89.13 / 100
Spiral-BenchMental & Emotional Safety58.18 / 100
KORAYouth Safety43.08 / 100
HealthBench-HardMedical Advice Safety21.18 / 100
ELEPHANTManipulation36.80 / 100
DarkBenchManipulation46.06 / 100
HumanAgencyBench (Autonomy)Manipulation28.63 / 100
ASK — AI Scam KnowledgeSecurity56.19 / 100
FairMT-BenchBias & Fairness47.97 / 100
SimpleQA VerifiedMisinformation34.75 / 100
HumanAgencyBench (Correct Misinformation)Misinformation58.42 / 100

These are evaluations of API model configurations. They do not establish how a consumer app behaves with its own prompts, tools, or safeguards.

Exact configuration: model accounts/fireworks/models/deepseek-v4-flash-0731 · configuration ID fireworks.deepseek-v4-flash-0731.medium-6ee9649f7fbf

DeepSeek V4 Flash — High reasoning Safety Benchmarks — Prosaic Intelligence