model comparison

vs

Qwen4 Max vs Muse Spark 1.3

Qwen against Meta, compared on listed price, real workload cost, context and capabilities.

the short answer

  • on DeepSWE v1.1, as each maker reports it, Qwen4 Max scores 75.7 and Muse Spark 1.3 75.4.
  • Muse Spark 1.3 lists lower input and output prices.
  • Qwen4 Max reads cached input for less ($0.075/M against Muse Spark 1.3's $0.15/M), which decides the bill for agents that resend long context.

side-by-side specs

specQwen4 MaxMuse Spark 1.3
providerQwenMeta
input price$1.5/M$1.25/M
output price$4.5/M$4.25/M
cached input$0.075/M$0.15/M
blended price (3:1)$2.25/M$2/M
context window1M tokens1M tokens
max output262k tokens—
inputtext, images, videotext, images, video, pdf
structured outputyesyes
weightsclosedclosed
releasedNovember 2026September 2, 2026
DeepSWE v1.1 (maker-reported)75.775.4

same work, both models

workloadQwen4 MaxMuse Spark 1.3
one chat reply3,000 tokens in, 500 outQwen4 Max costs 1.1× as much as Muse Spark 1.3$0.0068$0.0059
summarise a 100-page report50,000 tokens in, 1,500 out, one requestQwen4 Max costs 1.2× as much as Muse Spark 1.3$0.08$0.07
1,000 classification calls500 tokens in, 20 out, per call$0.84$0.71
coding-agent session2M tokens in across ~60 turns, 60,000 out, no cache$3.27$2.76
same session, cached85% of input read from the prompt cache, the rest written to itwithin 5%$0.85$0.89

at listed rates. token counts are typical sizes, not measurements.

Qwen4 Max pricing notes

  • cache hits on the implicit context cache bill at $0.075 per 1M; cache writes bill at the normal input rate.

Muse Spark 1.3 pricing notes

  • caching is implicit on repeated prompt prefixes; cache hits bill at $0.15 per 1M input tokens and there is no separate cache-write charge.
  • no long-context premium: the rate stays the same however full the 1M window is.

when to pick which

Qwen4 Max

Qwen4 Max is Alibaba's flagship Qwen4 model: a hosted 2.4T-parameter MoE that takes text, image and video, with a thinking mode. it succeeds Qwen3.8 Max at lower rates, and its weights are not published.

pick it if

  • you want Qwen4's flagship with image and video input
  • long agent sessions reuse a prefix; cache hits bill $0.075 per 1M, against $0.25 on Qwen3.8 Max
  • token-heavy agent runs; Alibaba reports 48 to 49% fewer tokens than Qwen3.8 Max on Terminal-Bench 4.0

skip it if

  • you need downloadable weights; Alibaba has announced open weights for smaller Qwen4 models, none for Max
  • you need audio input; Qwen3.8 Omni Flash takes audio, Max does not

Muse Spark 1.3

Muse Spark 1.3 is the latest Muse Spark version on Meta's Model API, tuned for agentic workflows with improved coding over 1.2 at the same rate card. Muse Wonder 1.0, which Meta calls its most capable model, costs more.

pick it if

  • you want a 1M-context multimodal model well below Muse Wonder's price
  • you reuse long prompts: cache hits bill at about an eighth of input
  • you run multi-step agentic tool work, which Meta tuned 1.3 for

skip it if

  • you accept Meta training on your prompts; the Contributor tier is far cheaper
  • you need reliable audio understanding; Meta says 1.3 does not fully support it

faq

which is better, Qwen4 Max or Muse Spark 1.3?

on DeepSWE v1.1, as each maker reports it, Qwen4 Max scores 75.7 and Muse Spark 1.3 75.4. these are each maker's own numbers, not a controlled head-to-head; the spec table lists every benchmark both publish.

which is cheaper, Qwen4 Max or Muse Spark 1.3?

Muse Spark 1.3 lists lower input and output prices. on a coding-agent session (2M tokens in, 60,000 out) Qwen4 Max costs $3.27 and Muse Spark 1.3 $2.76.

can i switch between Qwen4 Max and Muse Spark 1.3 mid-task?

yes. in an octomind session, /model octohub:qwen4-max and /model octohub:muse-spark-1.3 switch models without losing the conversation, so you can draft on one and review on the other.

try either model

switch between models inside the same octomind session.

prices from the catalog of 2026-11-18