DeepSeek V4.1 Flash 1017 vs DeepSeek V4.1 Flash: two coding tests

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

Short answer: DeepSeek V4.1 Flash 1017 and DeepSeek V4.1 Flash have the same per-token prices, inputs and 1M token context on EmpirioLabs. In our two one-shot coding tests, 1017 used fewer output tokens and less time on one task and more on the other, and both versions returned working animations on both. Use deepseek-v4-1-flash-1017 to pin DeepSeek's October release. We are not declaring a winner.

Watch the comparison

The video plays the same one-shot results shown below, rendered live. The music was generated by ACE-Step on EmpirioLabs. Watch it on YouTube.

Test 1: aquarium

Prompt: "Build an aquarium tank with five different kinds of fish swimming at different speeds, rising bubbles, swaying seaweed and light rays moving across the sand."

Test 2: city skyline

Prompt: "Build a city skyline at night where office windows light up and switch off at random, a train crosses a bridge in front of the buildings, and a full moon rises slowly behind them."

Specs and results at a glance

How we ran it

Both models received the identical prompt for each task, with the same closing instruction: return one complete, self-contained HTML file with inline CSS and JavaScript, no external libraries or assets, that animates on its own as soon as it loads. Both ran with default settings, thinking on at its default budget, and a 65,536 token output limit, one shot each, and the two models ran at the same time for each task. Each screenshot is the returned file rendered as-is in a headless browser after six seconds. Times are wall-clock seconds for the whole request on October 18, 2026, and token counts come from the API's usage report.

What to look for

Tokens and time. On the aquarium, 1017 finished in 124 seconds against 162 and used 21 percent fewer output tokens. On the skyline, 1017 reasoned longer, 42,907 output tokens against 37,928, and took 177 seconds against 163, though its finished file was shorter. Both versions bill the same rates, so the price of each task follows the tokens it used.

The scenes. Both versions put every requested element on screen in both tasks, and all four files animate on their own. Compare the fish designs, the bridge and train, and how each scene moves.

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": "deepseek-v4-1-flash-1017",
    "max_tokens": 65536,
    "messages": [{"role": "user", "content": "Build an aquarium tank with five different kinds of fish swimming at different speeds, rising bubbles, swaying seaweed and light rays moving across the sand. Return one complete, self-contained HTML file with inline CSS and JavaScript, no external libraries or assets, that animates on its own as soon as it loads."}]
  }'

To run the same request against the September release, swap model to deepseek-v4-1-flash before October 27, 2026. From that date deepseek-v4-1-flash serves the 1017 release, so both IDs return the same model. Try it in the playground.

Frequently asked questions

What changed in DeepSeek V4.1 Flash 1017?

It is DeepSeek's October 17 update to V4.1 Flash. DeepSeek reports stronger coding and agent results, faster task completion and shorter answers. On EmpirioLabs the two versions have the same prices, inputs and context window.

Were the results edited or retried?

No. Each screenshot is the first file the model returned for that prompt, rendered exactly as returned.

Is 1017 always cheaper per task?

Not in these tests. Both versions bill the same rates, so a task costs what its tokens cost: 1017 used fewer output tokens on the aquarium and more on the skyline.

Try it

Playground | All models | Pricing