DeepSeek V4.1 Flash Intermediate Build Opens Internal Testing with Native Multimodality

On September 8, multiple sources revealed that DeepSeek V4.1 Flash has entered internal testing as an "interim version," with core upgrades including a brand-new model architecture and native multimodal support. Officially, it is billed as more capable, faster, and cheaper.

Confirmed

• Synced reports that DeepSeek officially announced the V4.1 Flash interim version for internal testing, with major upgrades including a new model architecture, native multimodality, and stronger capabilities at higher speed and lower cost.
• How to use it: keep the baseurl unchanged and set the model name to an ID starting with deepseek-v4.1-flash-expires-on-09.
• According to an official WeChat group notice relayed by vista8, billing is the same as V4 Flash, with up to 20 concurrent requests per account.

Unconfirmed

• The news initially broke via bloggers such as op7418 and AGI Hunt from DeepSeek's official WeChat group, all stressing it lacked confirmation from an official public announcement; Synced later reported it as "officially announced," but a formal public announcement and full capability details are still pending.
• teortaxesTex relayed a community joke hoping it can at least beat GLM 5.3 Flash beyond vision—this is only community opinion with no benchmark evidence.

Why it matters

• This is the first time DeepSeek has been reported to release an internal test as an "interim version (expires-on)," with billing and concurrency rules clarified in tandem, reflecting a product cadence of iterating through real developer testing before official release.
• Native multimodality marks a major upgrade to DeepSeek's flagship model roadmap, and the community has already started benchmarking it against competitors like GLM 5.3 Flash—worth watching for real-world performance once the official version ships.

2026-09-08 ~ 2026-09-08 · 9 related posts

• How to call DeepSeek V4.1 Flash beta: model name and rate limits revealed — 赛博禅心 · 2026-09-08
• Rumored DeepSeek V4.1 Flash surfaces, user hopes it beats GLM 5.3 Flash — teortaxesTex · 2026-09-08
• Rumor: DeepSeek V4.1 Flash released with stronger native multimodal — op7418 · 2026-09-08
• [source] DeepSeek V4.1 Flash reportedly in private beta with new architecture and native multimodal support — vista8 · 2026-09-08
• [source] DeepSeek spotted testing V4-Flash-Vision: new arch, native multimodal, same price — teortaxesTex · 2026-09-08
• DeepSeek quietly tests V4.1 Flash API beta with new architecture and native multimodality — tokenbender · 2026-09-08

3 near-duplicate retellings: AGI Hunt · jiqizhixin · aigclink