Home Blog

Kimi K3.1 Pro vs Kimi K3 vs MiniMax M3.1 Flash vs DeepSeek V4 Pro 0813: four coding tests

Kimi K3.1 Pro vs Kimi K3 vs MiniMax M3.1 Flash vs DeepSeek V4 Pro 0813

Nov 5, 2026

EmpirioLabs AI

Disclosure: This article was written with AI assistance and reviewed by EmpirioLabs AI.

Short answer: we gave Kimi K3.1 Pro, Kimi K3, MiniMax M3.1 Flash and DeepSeek V4 Pro 0813 the same four prompts for animated scenes in a single HTML file. Kimi K3.1 Pro, MiniMax M3.1 Flash and DeepSeek V4 Pro 0813 ran at reasoning effort max on all four; Kimi K3 ran at high on three and at max on the drone light show. Kimi K3.1 Pro returned all four files in 617 to 632 seconds. Fifteen of the sixteen files animate on their own; one stops at a script error and stays black. All four models are available on EmpirioLabs through the API and the Playground. We are not declaring a winner: watch the results and compare.

Watch the comparison

Each model writes its code at its measured speed, then its file runs. The music was generated by ACE-Step on EmpirioLabs.

Specs at a glance

Kimi K3.1 ProKimi K3MiniMax M3.1 FlashDeepSeek V4 Pro 0813
ReleasedOctober 23, 2026July 15, 2026October 19, 2026August 13, 2026
Context window1,048,576 tokens1,048,576 tokens1,048,576 tokens1,000,000 tokens
InputText and imagesText, images and videoText, images and videoText
Maximum output131,072 tokens131,072 tokens524,288 tokens393,216 tokens
Output token priceLower than Kimi K3Highest of the fourLowest of the fourSecond lowest of the four
Setting in this testReasoning effort maxReasoning effort high, max on Test 4Reasoning effort maxReasoning effort max

Rates are on the pricing page and on each model's page.

Test 1: roller coaster at sunset

Prompt: "Create a single self-contained HTML file (all CSS and JavaScript inline, no external libraries, no CDN links, no imports) that animates a roller coaster at sunset on a full-window <canvas>. Requirements: a red train of four cars runs on a white steel track with hills, a tall drop and a full vertical loop, climbing slowly on the lift hill and speeding up down the drop, each car following the track and tilting with its slope, with the wooden supports, a glowing sky and distant hills drawn with depth, and riders' arms going up on the drop. Smooth 60fps animation that runs on its own and loops forever."

ModelReasoning effortTimeOutput tokensLines in the file
Kimi K3.1 Promax617 seconds28,083841
Kimi K3high718 seconds24,594263
MiniMax M3.1 Flashmax268 seconds38,950930
DeepSeek V4 Pro 0813max689 seconds38,727417

Test 2: wooden marble run

Prompt: "Create a single self-contained HTML file (all CSS and JavaScript inline, no external libraries, no CDN links, no imports) that animates a wooden marble run on a full-window <canvas>. Requirements: glass marbles in bright colors roll down zigzag ramps with real gravity and bounces, drop through a spinning wheel and a funnel that swirls them around before they fall through, knock a row of small dominoes on the way, and ride a chain lift back to the top. Each marble has a highlight and a soft shadow. Smooth 60fps animation that runs on its own and loops forever."

ModelReasoning effortTimeOutput tokensLines in the file
Kimi K3.1 Promax619 seconds29,9081,205
Kimi K3high312 seconds12,243373
MiniMax M3.1 Flashmax818 seconds53,7861,584
DeepSeek V4 Pro 0813max529 seconds36,000638

Test 3: inside a clock tower

Prompt: "Create a single self-contained HTML file (all CSS and JavaScript inline, no external libraries, no CDN links, no imports) that animates the inside of a mechanical clock tower on a full-window <canvas>. Requirements: large brass gears of different sizes mesh and turn at the correct ratios, a swinging pendulum drives an escapement that ticks, moonlight falls through the round clock face from behind, and when the minute hand reaches the top a bell swings and pigeons fly past the window. Smooth 60fps animation that runs on its own and loops forever."

ModelReasoning effortTimeOutput tokensLines in the file
Kimi K3.1 Promax626 seconds29,0851,098
Kimi K3high202 seconds8,229463
MiniMax M3.1 Flashmax587 seconds45,0561,497
DeepSeek V4 Pro 0813max852 seconds59,273818

Test 4: drone light show over a harbor

Prompt: "Create a single self-contained HTML file (all CSS and JavaScript inline, no external libraries, no CDN links, no imports) that animates a drone light show over a harbor at night on a full-window <canvas>. Requirements: hundreds of glowing drones lift off from a barge and fly as one swarm, morphing smoothly through a star, a heart, a butterfly that flaps its wings and a rocket, each drone easing to its own place and changing color with the shape, above a dark city skyline with lit windows, while the harbor water reflects the lights in rippling streaks. Smooth 60fps animation that runs on its own and loops forever."

ModelReasoning effortTimeOutput tokensLines in the file
Kimi K3.1 Promax632 seconds30,191939
Kimi K3max735 seconds30,402590
MiniMax M3.1 Flashmax305 seconds46,1551,320
DeepSeek V4 Pro 0813max281 seconds19,569546

MiniMax M3.1 Flash's file stops at a script error: line 171 reads formations before the constant is set on line 458, so the page stays black. That is the file as returned, unedited.

How we ran it

Each model received the identical prompt for each task, ending with the same instruction to return only the complete HTML file in one code block. Kimi K3.1 Pro, MiniMax M3.1 Flash and DeepSeek V4 Pro 0813 ran at reasoning_effort max. Kimi K3 ran at high on the roller coaster, marble run and clock tower, and at max on the drone light show. Every request streamed, with a 65,536 token output limit, on November 5, 2026. Times are wall-clock seconds for the whole request, and token counts come from the API's usage report. Each file is the first one the model returned for that prompt, recorded running in a headless browser exactly as returned; a request that returned no file was sent again.

What to look for

Time. Kimi K3.1 Pro took 617 to 632 seconds on every test, the narrowest spread of the four. MiniMax M3.1 Flash took 268 to 818 seconds and DeepSeek V4 Pro 0813 281 to 852. Kimi K3's 202 to 718 seconds at high are not comparable with the max runs; its max run on the drone light show took 735 seconds.

Output speed. Kimi K3.1 Pro wrote 45.5 to 48.4 tokens per second, Kimi K3 34.2 to 41.4, MiniMax M3.1 Flash 65.7 to 151.6 and DeepSeek V4 Pro 0813 56.2 to 69.6.

File length. MiniMax M3.1 Flash wrote the longest file in all four tests (930 to 1,584 lines) and Kimi K3.1 Pro the second longest (841 to 1,205). Kimi K3's shorter files at high (263 to 463 lines) are not a like-for-like comparison. File length is not quality: the longest file is not always the scene that moves best.

The scenes. Compare each model's lift hill and loop, ramps and funnel, gears and pigeons, and swarm shapes. DeepSeek V4 Pro 0813's track crosses itself, and its backdrop stops short of the window edges. Its marble run shows one or two marbles at a time where the others keep several moving. Kimi K3's clock tower at high is the smallest and dimmest of the four. Kimi K3.1 Pro's drones wait on the barge for about three seconds before they lift off.

Run the same test on EmpirioLabs

curl https://api.empiriolabs.ai/v1/chat/completions \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k3-1-pro",
    "reasoning_effort": "max",
    "max_tokens": 65536,
    "stream": true,
    "messages": [{"role": "user", "content": "Create a single self-contained HTML file (all CSS and JavaScript inline, no external libraries, no CDN links, no imports) that animates a wooden marble run on a full-window <canvas>. Requirements: glass marbles in bright colors roll down zigzag ramps with real gravity and bounces, drop through a spinning wheel and a funnel that swirls them around before they fall through, knock a row of small dominoes on the way, and ride a chain lift back to the top. Each marble has a highlight and a soft shadow. Smooth 60fps animation that runs on its own and loops forever. Return ONLY the complete HTML file inside a single ```html code block."}]
  }'

Swap model to kimi-k3, minimax-m3-1-flash or deepseek-v4-pro-0813 to run the same request on the others. A request this long takes minutes, so it streams the file as it is written. Or try them in the Playground.

Frequently asked questions

Were the results edited or retried?

No file was edited. Each file is the first one the model returned for that prompt, rendered exactly as returned. A request that returned no file was sent again.

Which setting did each model use?

Kimi K3.1 Pro, MiniMax M3.1 Flash and DeepSeek V4 Pro 0813 ran at reasoning_effort max on all four tests. Kimi K3 ran at high on the roller coaster, marble run and clock tower, and at max on the drone light show.

Why is one drone light show black?

MiniMax M3.1 Flash's file has a script error, so the animation never starts. It is shown as returned, like every other result.

Which model has the lowest rates?

MiniMax M3.1 Flash has the lowest input and output rates of the four and Kimi K3 the highest; Kimi K3.1 Pro is lower than Kimi K3. Each request costs what its tokens cost at the model's rates, so a model that writes a shorter file or reasons less costs less. See the pricing page for the numbers.

What is new in Kimi K3.1 Pro?

It is Moonshot's flagship model for long coding and agent work, released on October 23, 2026, with a 1M token context, image input, four reasoning effort levels, built-in web search and a code interpreter. See How to Use the Kimi K3.1 Pro API.

Try it

Playground | All models | Pricing

Ready to use better endpoints?

Explore our models, or contact us about business inquiries, custom deployments, or anything else.