[KongchangAI]
ModelDeepSeek V4.1 Flash

V4.1 Flash

An online service model version from DeepSeek whose specific details are not publicly disclosed.

Core Facts

Timeline (last 90 days)

Nov 14

DeepSeek released its full-suite application API this week, and its V4.1 Flash model already supports capabilities including web search, deep thinking, file input, link reading, and model safety under multiple interfaces.

Unverified50%
Nov 14

The time per task (average time required to complete a task) measured by Artificial Analysis shows that V4.1 Flash is 15.2 minutes, ranking second among these frontier models; GPT-6 Astra takes 24.7 minutes, Google's Gemline 4 Carbon takes 25.7 minutes, and Cloud Fable 5.2 takes 47.5 minutes.

Unverified50%
Nov 13

For tasks that require executing concrete actions, calling tools, and multi-step operations, you need to use an Agent vehicle like Hornets together with the latest model V4.1 Flash.

Unverified50%
Nov 13

Both the web version and the client of Hornets can use the latest large model, V4.1 Flash.

Unverified50%
Oct 23

DeepSeek updated V4.1 Flash to the 2017 version and debuted the Reflection feature.

Unverified50%
Oct 22

Whether V4.1 Flash has been officially released and what its specific version number is have not yet been confirmed by the official side.

Expired50%
Oct 22

The suspected V4.1 Flash, gray-tested on October 17, implemented the marquee selection feature for the same prompt "very perfectly", correctly handling the order of click-versus-drag determination.

Expired50%
Oct 22

The old model's implementation of the marquee selection feature had a bug: left-clicking first selected the object, and after dragging, the selection was canceled before entering marquee selection.

Expired50%
Oct 21

DeepInfer currently supports only one model, V4.1 Flash, and does not support Hopper and earlier GPUs.

Expired50%
Oct 21

The V4.1 Flash model contains 384 experts, only a few of which are activated per token; these experts are distributed across 64 cards, 6 per card, plus a clone of one popular chef.

Unverified50%

40 more timeline events

All Facts (20)

Verified

V4.1 Flash is natively multimodal in design, capable of directly viewing and understanding image content.

75%
Verified

This update adds support for the official model ID deepseek-flash; new sessions default to V4.1 Flash, but explicitly configured models take higher priority.

65%
Verified

DeepSeek V4.1 Flash reduces per-token storage from 3,500 bytes in the previous generation to approximately 890 bytes through KV cache inter-layer alternation and reuse optimization.

65%
Verified

Interfaces that previously called DeepSeek V4 Pro will automatically switch to the new V4.1 Flash model, requiring no code changes from developers.

65%
Verified

V4.1 Flash activates only 8 billion parameters during the input stage and 16 billion parameters during the output stage.

65%
Verified

DeepSeek open-sourced the inference framework DeepInfer on October 20, providing optimized operators, multi-card parallelism and asynchronous scheduling capabilities for V4.1 Flash, under the Apache 2.0 license.

80%
Verified

At 12:00 noon on September 14, V4 Pro will officially be taken offline, and all requests will be fully routed to V4.1 Flash.

70%
Unverified

The DeepSeek API offers three effort levels—low, high (default) and max—which the V4.1 Flash paper maps to internal effort values of 50, 75 and 100, and reports that going from high to max lengthens agent trajectories by 1.6–1.8 times while yielding only marginal gains.

60%
Unverified

The V4.1 Flash model weights have been uploaded to Hugging Face under a permissive license.

60%
Unverified

V4.1 Flash spent 45 minutes on a robotic arm visual control task without succeeding, with the bottleneck primarily in visual processing.

60%
Unverified

V4.1 Flash reduces KVCache to one-quarter of the previous generation.

60%
Unverified

The time per task (average time required to complete a task) measured by Artificial Analysis shows that V4.1 Flash is 15.2 minutes, ranking second among these frontier models; GPT-6 Astra takes 24.7 minutes, Google's Gemline 4 Carbon takes 25.7 minutes, and Cloud Fable 5.2 takes 47.5 minutes.

50%
Unverified

For tasks that require executing concrete actions, calling tools, and multi-step operations, you need to use an Agent vehicle like Hornets together with the latest model V4.1 Flash.

50%
Unverified

Both the web version and the client of Hornets can use the latest large model, V4.1 Flash.

50%
Unverified

The V4.1 Flash model contains 384 experts, only a few of which are activated per token; these experts are distributed across 64 cards, 6 per card, plus a clone of one popular chef.

50%
Unverified

V4.1 Flash's expert scheduling strategy: load is monitored every 1,000 steps, experts are reallocated every 10,000 steps, and ultimately the busiest card is only 9% busier than the average level.

50%
Unverified

September's V4.1 Flash consumed nearly twice as many tokens as open-source models of the same scale when running the same benchmark.

50%
Unverified

The September version's promotional page said '552B Backbone Parameters', emphasizing backbone parameters; this time the basis was switched from backbone parameters to all weights, and the architecture itself is unchanged.

50%
Unverified

V4.1 Flash's software engineering score of 74.2 is the highest in the comparison table, but leads the strongest flagship competitor by only 0.2 points.

50%
Unverified

The original three modes — "Fast / Expert / Real-time" — on DeepSeek's web and app clients have been merged and unified into V4.1 Flash, so users no longer need to select manually.

50%

Source Articles