Models / DeepSeek

DeepSeek V4 Flash 0423

by DeepSeek · deepseek/deepseek-v4-flash

DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. It is designed for fast inference and...

reasoningtool-usereleased 2026-04-24

JSON Route to this model

Legal score · raw

not yet benchmarked

Context

1.05M

max output 384K

Input

$0.08

per 1M tokens

Output

$0.16

per 1M tokens

Suite cost

run the suite to see

Benchmark results

Task	Category	Raw	Juiced	Correct	Latency	Cost	Ran
Hearsay Identification	Evidence	-	-	-	-	-	-
Bluebook Citation Format	Research & Writing	-	-	-	-	-	-
Federal Civil Procedure	Procedure	-	-	-	-	-	-
Limitations Arithmetic	Procedure	-	-	-	-	-	-
Contract Clause Classification	Contracts	-	-	-	-	-	-
Citation Hallucination Resistance	Reliability	-	-	-	-	-	-

Measured by DocketBuster

These numbers come from DocketBuster's own legal battery, not from DocketRouter's suite. Latest run per metric, with n and a 95% Wilson interval where the source reports one. See docketbuster.com/benchmarks.

Metric	Value	n	Interval	Measured
Statute pinpoint, exact section (no retrieval)	12.3% (37/300)	300	95% CI 9.1% to 16.5%	2026-08-21
Statute pinpoint, exact section (with DocketBuster retrieval)	76.5% (235/307)	307	95% CI 71.5% to 80.9%	2026-08-24
Say-nothing rate (declines to bluff when the answer is not in the record)	100.0%	300	95% CI 98.7% to 100.0%	2026-08-21
Abstained on statute pinpoint	16.3% (50/307)	307	count, no interval reported	2026-08-24
Coaching quality (GW-14x, 0 to 8)	7.16 / 8	-	rubric mean, no interval reported	2026-08-22

Source files: hard-llm-ortier-dsflash.json, gw14x-ortier-dsflash.json, hard-llm-level-dsflash-statute_rag.json. Raw model name in source: deepseek/deepseek-v4-flash.