Models
|
Benchmarks
|
Organizations
|
Recent Records
|
Sources
|
Documentation
DeepSeek V4-Flash on MMLU-Pro
BENCHMARK RECORD: BR-008V4FLASH-001
Canonical URL: https://www.benchmarkregistry.org/records/BR-008V4FLASH-001
Benchmark scores may depend on benchmark version, evaluation configuration, model snapshot, and evaluator. Records sharing a benchmark name are not necessarily directly comparable.
Record
Record Identifier BR-008V4FLASH-001Record Status ACTIVE Created At 2026-07-12T17:33:24.524Z Updated At 2026-07-12T17:33:24.524Z Superseded By Not applicable
Model
Official Model Name DeepSeek V4-Flash Model Identifier DPSK-V4FLASHOrganization DeepSeek Model Family V4FLASH Model Number Unknown Tier Unknown Model Snapshot Unknown Snapshot Label Unknown Provider Model Identifier UnknownSnapshot Date Unknown
Benchmark
Benchmark Family MMLU-Pro Benchmark Version or Variant Unknown Benchmark Version Reference mmlu-pro/defaultBenchmark Organization University of Waterloo (TIGER-Lab) Benchmark Status ACTIVE
Result
Metric Exact Match Score 86.2 Numeric Score 86.2 Evaluation Date Unknown Reported Date Unknown
Evaluation Context
Configuration Reference Unspecified Shots Unknown Reasoning Mode Unknown Reasoning Effort Unknown Pass Count Unknown Agent Scaffold Unknown Evaluation Harness Unknown Temperature Unknown Top P Unknown Max Output Tokens Unknown System Prompt Description Unknown Additional Configuration None reported
Source and Reporting
Primary Source DeepSeek-V4-Flash Model Card ↗ Source Type MODEL_CARD Publisher DeepSeek Source Published Date Unknown Source Accessed At 2026-07-12T17:30:50.104Z Report Type PROVIDER Evaluator DeepSeek Evaluator Type MODEL_PROVIDER
Provenance
Record Status
This record is ACTIVE . Registry identifiers remain reserved when record status changes.