DeepSeek releases V4-Flash-0731 with 304B parameters and enhanced agentic capabilities
The 304-billion-parameter model is ranked ahead of a 428B competitor on Artificial Analysis’ Intelligence Index and priced at $0.14 per million input tokens and $0.27 per million output tokens.
1 source · cross-referenced
- DeepSeek’s V4-Flash-0731 is a 304-billion-parameter model with substantially enhanced agentic capabilities.
- Artificial Analysis ranks it ahead of MiniMax M3 (428B) on its Intelligence Index.
- Pricing is $0.14 per million input tokens and $0.27 per million output tokens.
- User tests show improved image generation quality when reasoning effort is increased from default to high.
DeepSeek has released DeepSeek-V4-Flash-0731, a 304-billion-parameter model described as having substantially enhanced agentic capabilities.
According to Artificial Analysis, the model is ranked ahead of MiniMax M3—a 428-billion-parameter model—on its Intelligence Index.
Pricing for the model is listed at $0.14 per million input tokens and $0.27 per million output tokens.
User testing via OpenRouter showed that increasing the reasoning effort from default to high produced a marked improvement in generated image quality, transitioning from a flawed pelican illustration to a coherent scene of a pelican riding a bicycle.
- Jul 31, 2026 · OpenAI — News
OpenAI disrupts scam operation using ChatGPT in Cambodia
Trust74 - Jul 29, 2026 · OpenAI — News
OpenAI offers free advanced ChatGPT access to 100,000 academic researchers
Trust79 - Jul 29, 2026 · Google DeepMind — Blog
Google DeepMind launches Lyria 3.5 for Flow Music with improved musicality, lyrics, and vocals
Trust84