DeepSeek releases DeepSeek-V4.1-Flash, a 552B MoE with vision and MIT-licensed weights

Archive entry: this covers an event from September 10, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On September 10, 2026, DeepSeek released DeepSeek-V4.1-Flash, the smallest model of its Causal Encoder-Decoder architecture family, with native visual understanding and MIT-licensed weights on Hugging Face, as described in its API docs news post.

What changed

• Architecture: a 552B-parameter MoE with 8B active parameters for input and 16B for output. Compared with the previous generation, its KV cache needs 1/4 of the HBM and 1/8 of the SSD storage.
• API: served as deepseek-flash with a 1M context, 384K maximum output, concurrency of 2500 and vision support. V4-Flash and V4-Flash-Vision-Exp were retired, and their names temporarily route to V4.1-Flash.
• Prices per million tokens (off-peak/peak): cache hit $0.003/$0.006, cache miss $0.15/$0.30, output $0.60/$1.20, effective from 04:00 UTC on September 10. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, Monday to Friday, excluding Chinese public holidays.
• Scores: among the scores DeepSeek listed are Terminal-Bench 2.1 (90.6), Terminal-Bench 4.0 (31.2), DeepSWE v1.1 (74.2), GPQA Diamond (90.9) and Codeforces (3471). It also said tests by multiple parties put V4.1-Flash ahead of V4-Pro on performance, cost, speed and total runtime.
• Open weights: published on Hugging Face as deepseek-ai/DeepSeek-V4.1-Flash under the MIT license, together with the DeepSeek_V41_Tech_Report.pdf.

What it means for people who build with AI

DeepSeek's default fast model came with native vision and a 1M context, cache-miss input cost $0.15 per million tokens off-peak, and the weights were open. Official partners WorkBuddy (including CodeBuddy) and OpenCode supported the model.

A note on V4 Pro routing

The news post said deepseek-v4-pro requests would route to V4.1-Flash from 04:00 UTC on September 14, 2026. The change log later said V4 Pro API service would continue after September 14 with billing unchanged. More details are on the Hugging Face model page and the pricing page.

Sources

• DeepSeek-V4.1-Flash: Smarter, Faster, More Efficient | DeepSeek API Docs (DeepSeek, September 10, 2026, official)
• deepseek-ai/DeepSeek-V4.1-Flash (DeepSeek, official)
• Change Log | DeepSeek API Docs (DeepSeek, September 10, 2026, official)
• Models & Pricing | DeepSeek API Docs (DeepSeek, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.