DeepSeek V4.1 Flash 1017 Video Gallery

Sky tracks vs GPT-6.1 Sol Ultra

The original creator compares two rendered scenes. The original “0.44 cents” conflicts with a repost’s dollar wording, so no converted cost conclusion is displayed.

Rocket launch vs GPT-6.1 Sol

Two rocket-launch scenes play side by side. Production times and costs are source-reported; the comparison model is retained without inferring general performance from one run.

Harness coding-task recording

A terminal recording of a coding task and output. The source speed, free-access period and capability comparisons are not treated as verified product specifications.

Honeybee on a sunflower

The creator reports one turn in DeepSeek Harness. The recording shows the sunflower, bee and camera movement; the full generation process is not supplied.

Web layout and transitions (repost)

The original video link points to @Fei2411; this post is a repost. Model attribution follows the reposter, with no verified production process; the broad claim of beating Opus is not adopted.

Original video: @Fei2411 ↗

Electric motor animation vs Opus 5.5

The original creator shows a 3D explanation of a Tesla-style electric motor. Both model labels remain; the same-prompt claim is source-reported and engineering accuracy is unverified.

Voxel Roadster stunt vs GPT-6 Luna

A voxel-scene comparison published by Atomic Chat, retaining both model panels and platform credit. This promotional showcase is not an independent test; costs are publisher-reported.

Space train vs GPT-6 Astra

The panels are labeled DeepSeek and GPT-6 Astra. Reported run times and costs differ; this recording does not establish an overall model ranking.

Blender and Three.js hardware short

The creator reports using the model with Blender and Three.js. The short shows a graphics card, computer case and internal components; it is not native video generation.

Forest campfire vs MiniMax

The clip compares forest campfire scenes and lighting. The creator reports the same prompt and task; token counts, costs and run times are not independently verified or used to infer general cost efficiency.

Dam collapse vs Opus 5.5

Both Opus 5.5 and DeepSeek labels remain visible. The creator reports the same prompt and one-shot runs; costs and elapsed times have not been independently reproduced.

Typography and graphic motion

The creator reports a one-shot run in DeepSeek Harness. The clip shows typography, graphics and curve animations; the prompt, execution log and cost are not independently verified.

Typography and easing-curve video

The publisher attributes this motion video to the model version. It shows typography, curves and a countdown; tools and full instructions are undisclosed, and the source model ranking is not adopted.

Four concurrent streams on two DGX Sparks

The recording shows four concurrent streams, a different workload from the earlier single-terminal demo. The creator reports two DGX Sparks; on-screen rates are not benchmark results reproduced by this site.

Blender rocket launch and lunar scene

The creator reports roughly five hours of development and sixteen hours of Blender Cycles rendering. The model participated in production; this is not native video generation.

Interactive gummy bears

The second clip is labeled DeepSeek; the first shows GLM 5.3 Flash and is omitted. The creator reports local runs with the same harness and Max effort.

Forest and lakeside level

A walkthrough of a forest, water and terrain scene. The creator names DeepSeek V4.1 Flash 1017; the production steps and full prompt are not disclosed.

OpenDesign API tutorial motion

The OpenDesign team reports using the model with OpenDesign Motion to create this API setup tutorial. The platform promotion and model role are both retained.

Miku SVG drawing process comparison

A side-by-side drawing process from DeepSeek and Gemini 4 Argon. The source prefers Argon for a human-like drawing sequence; this is a task-specific observation.

Rotating miniature planet

The publisher shows a low-poly planet with towns and forests, reporting one output followed by minor adjustments. Model attribution and process are publisher-reported.

Block world in one HTML file

The source describes a Minecraft-style world in a single HTML file. The recording shows movement, block selection and building; code is not included in the supplied data.

Koi pond and interaction controls

The creator reports using DeepSeek Harness and describes fish, ripples, feeding, currents and fish profiles. The pond and controls are visible; token totals and costs are source-reported, without billing or full interaction verification.

Horror game vs Opus 5.5

The creator reports building Five Nights at Freddy’s-style games with both models. Both labels remain; the recording does not establish full playability or licensing clearance.

Cycling pelican vs the previous Flash

The recording compares DeepSeek V4.1 Flash 1017 with the previous Flash. The source names DeepSeek via OpenRouter and OpenCode as the harness; time and cost are not independently reproduced.

Cloud-whale city walkthrough

The source names the model and calls the scene a cloud-whale city. The clip shows a city on a whale, clouds and camera movement; the full prompt and toolchain are unknown.

Leifeng Pagoda and environment controls

The clip shows the pagoda, lake, mountains and changing lighting. The publisher reports a single instruction; architectural fidelity and Opus comparisons remain source assessments.

Breaching whale vs Kimi K3

The negative observations are retained: the creator reports blown-out colors and a poor whale on the first attempt, then gives DeepSeek a budget before comparing it with Kimi K3. This is not an equal-turn one-shot test.

Local code output on two DGX Sparks

The creator reports running a quantized model on two DGX Sparks. The clip shows terminal code output; the subjective speed assessment is not treated as a benchmark.

FAWNBENCH deer vs MiniMax

Two deer-motion scenes retain DeepSeek and MiniMax labels. Time, cost and quality judgments are creator-reported; this demonstration does not establish an overall model ranking.

Gummy-bear physics (secondary report)

The original video link points to @teortaxesTex; this is a secondary showcase. Gummy-bear interactions are visible; model attribution, production time and cost follow the secondary report and remain unverified.

Original video: @teortaxesTex ↗

Four-stroke engine pixel diagram

The caption uses the shorthand “Deepseek Flash 1017” and the video label says DeepSeek V4 Flash. The animation shows intake, compression, power and exhaust. Attribution follows the source; engineering accuracy is not independently verified.

Jumpgate: three-model comparison

The clip compares hyperspace effects from DeepSeek V4.1 Flash 1017, GLM 5.3 Flash and Fable 5.1. The ranking is the creator’s opinion. The root post points to a later prompt; undisclosed instructions are not reconstructed.

Animation and web design vs MiniMax

Panels labeled MiniMax M3.1 Flash and DeepSeek V4.1 Flash 1017 show animated characters and web designs. The publisher’s promotional framing is retained; on-screen model labels do not verify the backend.

Terminal output on B200 hardware

The publisher shows terminal output and reports updating the Nvidia megakernel on B200 hardware for the Cultura Builder community. Hardware, kernel setup and speed are source-reported, without independent measurement.

Kamakurakokomae railway scene

The publisher shows a coastal railway crossing, train and changing viewpoints, reporting three prompts. Detail fidelity and claims of beating other models remain source assessments, rather than verified conclusions.

Four-model first-person parkour

Four panels compare DeepSeek, Haiku 5.5, GPT-6 Luna and Qwen3.8 Flash. The source specifies one HTML file, four stages and one fix round, rather than a no-revision output.

BuildingBench architecture comparison

The clip compares reference photos, two DeepSeek versions and GPT-6 Astra building models. Scores and costs are publisher-reported, rather than a new evaluation by this site.

Three.js robot assembly and launch

The creator reports a robot-assembly and launch test via AiHubMix. The clip shows the scene; physical realism and execution cost are not independently verified.

Tokyo Tower vs Sonnet 5.5

Side-by-side Three.js scenes. The creator reports nineteen revisions for DeepSeek and eleven for Sonnet; this is not an equal-turn one-shot test.

Cycling peacock and motion defects

The creator reports no recursive loop and notes leg bending, tail-feather issues and limited bike movement. This is the DeepSeek version, without mixing in the later Gemini clip.

Roadster doors and spoiler controls

The creator reports a moving spoiler, scissor doors and camera views, while acknowledging limited exterior fidelity. The recording shows interactions; full code is not supplied.

Interior scene vs Haiku 5.5

The recording compares interior scenes labeled DeepSeek and Claude Haiku; the caption names Haiku 5.5. Model identity and comparison conditions are not independently verified.