DeepSeek launches vision-enabled AI model and cuts weekend API pricing

DeepSeek has launched DeepSeek-V4-Flash-Vision-Exp, an experimental multimodal AI model that adds visual processing to its flagship architecture. While DeepSeek claims the model's agent performance closely rivals Anthropic's Opus-4.8, benchmark data shows it wins on only three of eleven tests, trailing significantly on repository-scale tasks. Commercially, DeepSeek is bolstering its appeal by eliminating weekend peak API pricing starting August 23, 2026, charging all weekend usage at off-peak rates.

DeepSeek V4-Flash-Vision-Exp launch

• ▪DeepSeek launched an experimental multimodal AI model named DeepSeek-V4-Flash-Vision-Exp on August 21, 2026, which adds visual comprehension to its flagship text-based V4 Flash architecture.
• ▪DeepSeek-V4-Flash-Vision-Exp is available to developers through the DeepSeek API platform and is supported out-of-the-box by version 0.1.1 of the DeepSeek agent harness.
• ▪DeepSeek-V4-Flash-Vision-Exp preserves the foundational text benchmarks of DeepSeek-V4-Flash, including its reasoning capacity, broad factual knowledge, and autonomous agent features.

Benchmark performance vs Anthropic Opus-4.8

• ▪DeepSeek-V4-Flash-Vision-Exp trailed Anthropic's Opus-4.8 on eight benchmarks, including a 12-point deficit on NL2Repo and an 8.1-point deficit on DSBench-Hard.
• ▪DeepSeek-V4-Flash-Vision-Exp outperformed Anthropic's Opus-4.8 model on three out of eleven benchmarks published by DeepSeek, specifically winning on DeepSWE, Agents' Last Exam, and ZeroBench.
• ▪Anthropic's Opus-4.8 model, released in May 2026, remains fully supported and recommended for use with no retirement scheduled earlier than May 2027, despite Anthropic launching Claude Opus 5 on July 24, 2026.

Multimodal agent capability improvements

• ▪DeepSeek-V4-Flash-Vision-Exp showed a major leap in multimodal agent benchmarks over the text-only V4-Flash, scoring 36.5 against 26.2 on ApexBench and 27.3 against 25.2 on Agents' Last Exam.
• ▪Across seven text benchmarks evaluated by DeepSeek, DeepSeek-V4-Flash-Vision-Exp outperformed the text-only V4-Flash model on six benchmarks, with Cybergym being the sole exception.
• ▪DeepSeek disclosed in a benchmark footnote that the text-based V4-Flash model was scored on tests containing images it could not see, which partially accounts for its lower scores on multimodal benchmarks.

Weekend API pricing changes

• ▪DeepSeek announced it will stop distinguishing between peak and off-peak API pricing on weekends, charging all Saturday and Sunday usage at off-peak rates starting August 23, 2026, at 12:00 a.m. Beijing time.
• ▪DeepSeek API fees incurred before August 23, 2026, will continue to be settled under the company's previous pricing standards.