Deepseek releases V4.1-Flash multimodal model
Deepseek released V4.1-Flash, a 552 billion parameter multimodal model that cuts KV cache memory to a quarter of its predecessor. It ships under the MIT license and narrowly beats Opus 5 and GPT-5.6 Sol on the DeepSWE coding benchmark.
First seen 9 Sep, 11:19 UTC on Hacker News6 sourcesLast update 39d ago
DeepSeekLaunch · 9 Sep
DeepSeek V4.1 Flash
Open-weight model
$0.15 / $0.60per M tokens, in / out
1Mcontext
Openweights
How DeepSeek V4.1 Flash compares
List prices per million tokens, context and Arena's text leaderboard, against DeepSeek's other top models and the leading labs' best. The percentages are DeepSeek V4.1 Flash's price against each.
| Model | Released | Input | Output | Context | Arena |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash | September 2026 | $0.15 | $0.60 | 1M | #34 1475 |
| Claude Opus 5.5 | September 2026 | $4.00 -96% | $20.00 -97% | 1M | #2 1507 |
| GPT-6 Astra | September 2026 | $10.00 -98% | $50.00 -99% | 1.05M | #30 1475 |
| Gemini 4 Argon | October 2026 | $2.00 -92% | $10.00 -94% | 1M | #1 1525 |
Arena ranks the best effort setting of each model among all models on its text leaderboard. Every model and price.
What the sources say
Linked, never rewritten. Official means the lab itself.
Official1
Hugging Face · deepseek-ai· 10 Sep, 02:17deepseek-ai/DeepSeek-V4.1-Flash published on Hugging FaceHow it unfolded
Every source in the order it appeared. Times in UTC.
9 Sep 11:19Hacker News DeepSeek launching v4.1 flash cheaper and more capable than v4 pro10 Sep 02:17Hugging Face · deepseek-ai deepseek-ai/DeepSeek-V4.1-Flash published on Hugging Face10 Sep 06:11Hacker News DeepSeek v4.1 Flash10 Sep 06:21OpenRouter DeepSeek: DeepSeek V4.1 Flash listed on OpenRouter10 Sep 12:40The Decoder New Deepseek model V4.1-Flash cuts memory needs for AI agents10 Sep 23:45SiliconANGLE DeepSeek releases V4.1-Flash, says it outperforms flagship V4-Pro
Keep up with DeepSeek
DeepSeek on the ScoreIts timelineFollow to mark DeepSeek's stories across the site. No account needed.