DeepSeek V4.1 Pro vs Muse Spark 1.2: Same Window, Opposite Modality

Two models at around two dollars per million input tokens, one of them with a much larger output ceiling — and a fork that no price column shows. Muse Spark 1.2 costs $1.25 per million input tokens and $4.25 output, serves a 1M context, and accepts text, image, video, file and audio as input. DeepSeek V4 Pro costs $0.66 and $1.98 off-peak, doubles in two weekday windows, and is text-only. DeepSeek V4.1 Pro — the name this page exists for — has no card at all.

So the comparison that can actually be run is Muse against V4 Pro, and it turns on one question: does anything in your input arrive as a picture or a file. If yes, the DeepSeek side is not cheaper, it is unavailable.

The cards

• Muse Spark 1.2 — $1.25 in, $4.25 out; 1M context; measured p50 first token 3.03 seconds; input text, image, video, file and audio; output text.

• DeepSeek V4 Pro — $0.66 in, $1.98 out off-peak ($1.32 and $3.96 in peak); $0.022 cached; 1M context; 384K output; text-only; 6.22 seconds to first token.

• DeepSeek V4.1 Pro — no card, no price, no endpoint. Named once in September.

The numbers, per task

One million input tokens and 100,000 output tokens. Muse: $1.25 plus $0.425 — $1.68. DeepSeek off-peak: $0.66 plus $0.198 — 86 cents. In a peak window: $1.72, which is within 3% of Muse's flat price.

That is the crisp version of the whole comparison: DeepSeek is half the price off-peak and the same price during seven hours of every weekday, while being twice as slow to first token and unable to read a screenshot. For a scheduler, DeepSeek wins. For anything with a human or an image in it, Muse does.

Muse is also 3.2 seconds faster to first token, which is the kind of gap that decides a product's feel rather than its bill.

Why the missing tier matters less here

In most DeepSeek matchups the absent V4.1 Pro is the story. Here it is not, because the fork is modality: even a shipped V4.1 Pro would probably inherit the V4 generation's shape, in which the Pro tier was the text-only one and Flash carried the vision. DeepSeek's own price sheet marks vision "Not supported" for Pro and a check mark for Flash. If that pattern holds into V4.1, the missing Pro tier would not fix the thing that separates these two models.

That is worth stating because it is the difference between waiting for a release and waiting for a redesign.

What to run

Muse Spark 1.2 and DeepSeek V4 Pro are both on OrcaRouter at their vendors' published rates, on one OpenAI-compatible endpoint with automatic failover. The routing rule writes itself: multimodal input to Muse, text-only volume to DeepSeek in its off-peak window.

The three checks for V4.1 Pro are unchanged — changelog, repository, price column — and none has moved as of 30 October. This page's argument, though, does not depend on them: the modality split is already decided by the cards that exist.