DeepSeek V4 Flash Stumbles on Real Agent Tasks as Prices Surge
The gap between V4 Flash's benchmark scores and its real-world agent performance, combined with the price increase, challenges the model's value proposition in enterprise environments where orchestration matters as much as raw capability.
Reporting from 1 source: GIGAZINE.
DeepSeek's V4 Flash model, popular for its low cost and high benchmark scores, is facing criticism after third-party tests showed it struggling on real-world agent tasks. Composio ran 30 difficult multi-step workflows across four harnesses, with only 6 fully succeeding. The company also raised API prices on August 6, 2026, citing unprecedented demand.
DeepSeek's V4 Flash has been a hit for its low price and strong benchmark results, but third-party testing suggests the model may not live up to its scores when put to work on real tasks. Composio, an AI agent platform, ran 30 difficult multi-step workflows across four agent harnesses, using tools like Gmail, GitHub, Slack, and Google Sheets. Out of 240 total runs, 129 passed, and only 6 of the 30 workflows succeeded across every harness.
The cost per successful task stayed under $0.20 even on the priciest setup, so efficiency remains decent. But Composio notes that DeepSeek's August 6 API price increase could change that calculus. The company has not explained the hike, though some attribute it to a surge in demand that has also slowed inference speeds.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.