Evidence
- Sep 6Mostly beats non-vision V4 Flash and is said to be roughly on par with Claude Opus 4.8
- Aug 23Experimental vision model released on the DeepSeek API as deepseek-v4-flash-vision-exp, accepting mixed text and image input
- Aug 23Reported a large gain over V4-Flash on multimodal agent benchmarks, close to Opus-4.8; text ability said to match V4-Flash
- Aug 21Images are billed at V4-Flash token rates, capped at 384 tokens each; Files API is free and supports reuse by file_id
- Aug 21Reported 59.3% on DeepSWE, within error of GPT 5.6 Sol Medium (61%) at likely far lower cost, per a commenter
Show all 7
- Aug 21Misread an analog clock (5:10:45) in a commenter's test where Qwen3.8 27B came nearly right
- Aug 21Read a three-series business chart correctly, pulled exact callout numbers and reasoned about leading indicators and regime shift