DeepSeek V4.1
Timeline (last 90 days)
These 8 fused operators achieved significant performance gains compared to the serial computation of small operators.
Using Campbot DSL, the team developed 8 fused operators for the DeepSeek V4.1 network, including index, prolog, kingram, and git, among others.
The Attention layer of DeepSeek V4.1 can be divided into three stages: pre-processing, Attention computation, and post-processing.
DeepSeek V4.1 claims to achieve 74 on SWE, but the blogger found that this was the best result cherry-picked from multiple tests, with open code testing at 65 and PI testing at 66.
This content comes from a personal hands-on demonstration video by a single UP host and has not yet been independently verified by a third party.
All three models (Sol, Luna, and DeepSeek) failed to accurately reproduce multi-image layouts, which was a common shortcoming.
In a typography replication test, GPT-6 Sol was the fastest, subjectively completing the task in about four to five minutes; Luna took approximately three times as long as Sol (close to twenty minutes); DeepSeek was the slowest at about half an hour.
In terms of token consumption, Sol was the lowest; Luna's consumption was slightly more than twice that of Sol; DeepSeek's was higher than Luna's.
After being asked to rework and fix the output, Sol completed the multi-image layout and added decorative text beside images, while Luna and DeepSeek produced the multi-image layout but did not add the decorative text.
In the style-preservation step after copying and pasting into the WeChat Official Account platform, Sol and DeepSeek retained styles largely intact, while Luna's styles became disordered at the end.
10 more timeline events
All Facts (16)
Liu Shengyu is a DeepSeek researcher who participated in writing the main Attention operator for DeepSeek V4.1.
60%UnverifiedThese 8 fused operators achieved significant performance gains compared to the serial computation of small operators.
50%UnverifiedThe Attention layer of DeepSeek V4.1 can be divided into three stages: pre-processing, Attention computation, and post-processing.
50%UnverifiedUsing Campbot DSL, the team developed 8 fused operators for the DeepSeek V4.1 network, including index, prolog, kingram, and git, among others.
50%UnverifiedDeepSeek V4.1 claims to achieve 74 on SWE, but the blogger found that this was the best result cherry-picked from multiple tests, with open code testing at 65 and PI testing at 66.
50%UnverifiedThis content comes from a personal hands-on demonstration video by a single UP host and has not yet been independently verified by a third party.
50%UnverifiedThe test conclusions come from a single tester focused on the specific scenario of WeChat Official Account typography replication; the scores are subjective and do not represent the models' full capabilities in dimensions such as programming, reasoning, or long-context handling.
50%UnverifiedIn the style-preservation step after copying and pasting into the WeChat Official Account platform, Sol and DeepSeek retained styles largely intact, while Luna's styles became disordered at the end.
50%UnverifiedIn a typography replication test, GPT-6 Sol was the fastest, subjectively completing the task in about four to five minutes; Luna took approximately three times as long as Sol (close to twenty minutes); DeepSeek was the slowest at about half an hour.
50%UnverifiedIn terms of token consumption, Sol was the lowest; Luna's consumption was slightly more than twice that of Sol; DeepSeek's was higher than Luna's.
50%UnverifiedAll three models (Sol, Luna, and DeepSeek) failed to accurately reproduce multi-image layouts, which was a common shortcoming.
50%UnverifiedAfter being asked to rework and fix the output, Sol completed the multi-image layout and added decorative text beside images, while Luna and DeepSeek produced the multi-image layout but did not add the decorative text.
50%UnverifiedThe thinking intensity levels set for the test were: Max for Sol and DeepSeek, and High for Luna.
50%UnverifiedThe MoE design with large parameters but small activations lowers the hardware barrier for local deployment.
50%UnverifiedDeepSeek V4.1 ranked first in tests including Terminal 2.1, software engineering, cybersecurity, office automation, and agent terminal examinations.
50%UnverifiedDeepSeek V4.1 introduces native dynamic vision capabilities, enabling it to understand and process real-time images on screen.
50%