DeepSeek V4 Pro 0813 vs Muse Spark Open

Two models head to head

This comparison pits two open-weight models that you can install on your own servers against each other: DeepSeek V4 Pro 0813 and Muse Spark Open, one from DeepSeek, the other from Meta. The first dates from August 13, 2026, the second from October 15, 2026.

All figures below come from public sources checked on October 22, 2026: models.dev and OpenRouter for prices and capabilities, Epoch AI for scores.

What the benchmarks say

On our capability index, DeepSeek V4 Pro 0813 scores 72.2 out of 100 and Muse Spark Open 75.2. Muse Spark Open leads by 3 points.

On the 4 benchmarks reported for both, DeepSeek V4 Pro 0813 wins 3 times and Muse Spark Open 1:

• GPQA Diamond (PhD-level science questions): DeepSeek V4 Pro 0813 91.7%, Muse Spark Open 88.6%
• OTIS Mock AIME (competition mathematics problems): DeepSeek V4 Pro 0813 98.6%, Muse Spark Open 98.1%
• FrontierMath (very hard research-level mathematics): DeepSeek V4 Pro 0813 64.6%, Muse Spark Open 69.8%
• APEX-Agents (professional tasks carried out by an agent): DeepSeek V4 Pro 0813 47.3%, Muse Spark Open 33.4%

Pricing

DeepSeek V4 Pro 0813 costs $0.66 per million input tokens and $1.98 per million output tokens; Muse Spark Open costs $1.25 and $4.25.

For a typical agent task (read 20,000 tokens, write 2,000), expect $0.017 with DeepSeek V4 Pro 0813 versus $0.034 with Muse Spark Open. DeepSeek V4 Pro 0813 is 2 times cheaper, a gap that adds up quickly over 1,000 tasks: $16.34 difference.

Run the numbers with your own volumes in the agent cost calculator.

Under the hood

Both models have the same context window: 1,048,576 tokens, roughly 1,573 pages.

Reasoning mode: DeepSeek V4 Pro 0813 yes, Muse Spark Open yes. Tool calling: DeepSeek V4 Pro 0813 yes, Muse Spark Open yes. Image input: DeepSeek V4 Pro 0813 no, Muse Spark Open yes.

Both publish their weights: you can host them yourself.

In short

Our reading of the data:

• Choose DeepSeek V4 Pro 0813 for a lower cost.
• Choose Muse Spark Open for a higher capability index, a more recent release and image input.

This verdict relies on public figures; it does not replace a trial on your own tasks.